TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III: GLOBAL RECONSTRUCTION ALGORITHMS Shinichi Mochizuki November 2015 Abstract. In the present paper, which forms the third part of a three-part series on an algorithmic approach to absolute anabelian geometry, we apply the ab- solute anabelian technique of Belyi cuspidalization developed in the second part, together with certain ideas contained in an earlier paper of the author concerning the category-theoretic representation of holomorphic structures via either the topologi- cal group SL 2 (R) or the use of “parallelograms, rectangles, and squares”, to develop a certain global formalism for certain hyperbolic orbicurves related to a once- punctured elliptic curve over a number field. This formalism allows one to construct certain canonical rigid integral structures, which we refer to as log-shells, that are obtained by applying the logarithm at various primes of a number field. More- over, although each of these local logarithms is “far from being an isomorphism” both in the sense that it fails to respect the ring structures involved and in the sense [cf. Frobenius morphisms in positive characteristic!] that it has the effect of exhibiting the “mass” represented by its domain as a “somewhat smaller collection of mass” than the “mass” represented by its codomain, this global formalism allows one to treat the logarithm operation as a global operation on a number field which satisfies the property of being an “isomomorphism up to an appropriate renormal- ization operation”, in a fashion that is reminiscent of the isomorphism induced on differentials by a Frobenius lifting, once one divides by p. More generally, if one thinks of number fields as corresponding to positive characteristic hyperbolic curves and of once-punctured elliptic curves on a number field as corresponding to nilpotent ordinary indigenous bundles on a positive characteristic hyperbolic curve, then many aspects of the theory developed in the present paper are reminiscent of [the positive characteristic portion of] p-adic Teichmüller theory. Contents: Introduction §0. Notations and Conventions §1. Galois-theoretic Reconstruction Algorithms §2. Archimedean Reconstruction Algorithms §3. Nonarchimedean Log-Frobenius Compatibility §4. Archimedean Log-Frobenius Compatibility §5. Global Log-Frobenius Compatibility Appendix: Complements on Complex Multiplication 2000 Mathematical Subject Classification. Primary 14H30; Secondary 14H25. Typeset by AMS-TEX 1 2 SHINICHI MOCHIZUKI Introduction §I1. Summary of Main Results §I2. Fundamental Naive Questions Concerning Anabelian Geometry §I3. Dismantling the Two Combinatorial Dimensions of a Ring §I4. Mono-anabelian Log-Frobenius Compatibility §I5. Analogy with p-adic Teichmüller Theory Acknowledgements §I1. Summary of Main Results Let k be a finite extension of the field Q p of p-adic numbers [for p a prime def number]; k an algebraic closure of k; G k = Gal(k/k). Then the starting point of the theory of the present paper lies in the elementary observation that although the p-adic logarithm × log k : k k [normalized so that p → 0] is not a ring homomorphism, it does satisfy the important property of being Galois-equivariant [i.e., G k -equivariant]. def In a similar vein, if F is an algebraic closure of a number field F , G F = Gal(F /F ), and k, k arise, respectively, as the completions of F , F at a nonar- chimedean prime of F , then although the map log k does not extend, in any natural × way, to a map F F [cf. Remark 5.4.1], it does extend to the “disjoint union of the log k ’s at all the nonarchimedean primes of F in a fashion that is Galois- equivariant [i.e., G F -equivariant] with respect to the natural action of G F on the × resulting disjoint unions of the various k k. Contemplation of the elementary observations made above led the author to the following point of view: The fundamental geometric framework in which the logarithm operation should be understood is not the ring-theoretic framework of scheme theory, but rather a geometric framework based solely on the abstract profinite groups G k , G F [i.e., the Galois groups involved], i.e., a framework which satisfies the key property of being “immune” to the operation of applying the logarithm. Such a group-theoretic geometric framework is precisely what is furnished by the enhancement of absolute anabelian geometry which we shall refer to as mono- anabelian geometry that is developed in the present paper. This enhancement may be thought of as a natural outgrowth of the algorithm- based approach to absolute anabelian geometry, which forms the unifying theme [cf. the Introductions to [Mzk20], [Mzk21]] of the three-part series of which the present paper constitutes the third, and final, part. From the point of view of the present paper, certain portions of the theory and results developed in earlier papers of the present series most notably, the theory of Belyi cuspidalizations developed in TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 3 [Mzk21], §3 are relevant to the theory of the present paper partly because of their logical necessity in the proofs, and partly because of their philosophical relevance [cf., especially, the discussion of “hidden endomorphisms” in the Introduction to [Mzk21]; the theory of [Mzk21], §2]. Note that a ring may be thought of as a mathematical object that consists of “two combinatorial dimensions”, corresponding to its additive structure, which we shall denote by the symbol , and its multiplicative structure, which we shall denote by the symbol  [cf. Remark 5.6.1, (i), for more details]. One way to understand the failure of the logarithm to be compatible with the ring structures involved is as a manifestation of the fact that the logarithm has the effect of “tinkering with, or dismantling, this two-dimensional structure”. Such a dismantling operation cannot be understood within the framework of ring [or scheme] theory. That is to say, it may only be understood from the point of view of a geometric framework that “lies essentially outside”, or “is neutral with respect to”, this two-dimensional structure [cf. the illustration of Remark 5.10.2, (iii)]. One important property of the p-adic logarithm log k discussed above is that the image log k (O k × ) k which is compact may be thought of as defining a sort of canonical rigid integral structure on k. In the present paper, we shall refer to the “canonical rigid integral structures” obtained in this way as log-shells. Note that the image log k (k × ) of k × via log k is, like log k (O k × ) [but unlike k × !], compact. That is to say, the operation of applying the p-adic logarithm may be thought of as a sort of “compression” operation that exhibits the “mass” represented by its domain as a “somewhat smaller collection of mass” than the “mass” represented by its codomain. In this sense, the p-adic logarithm is reminiscent of the Frobenius morphism in positive characteristic [cf. Remark 3.6.2 for more details]. In particular, this “compressing nature” of the p-adic logarithm may be thought of as being one that lies in sharp contrast with the nature of an étale morphism. This point of view is reminiscent of the discussion of the “fundamental dichotomy” between “Frobenius- like” and “étale-like” structures in the Introduction of [Mzk16]. In the classical p-adic theory, the notion of a Frobenius lifting [cf. the theory of [Mzk1], [Mzk4]] may be thought of as forming a bridge between the two sides of this dichotomy [cf. the discussion of mono-theta environments in the Introduction to [Mzk18]!] that is to say, a Frobenius lifting is, on the one hand, literally a lifting of the Frobenius morphism in positive characteristic and, on the other hand, tends to satisfy the property of being étale in characteristic zero, i.e., of inducing an isomorphism on differentials, once one divides by p. In a word, the theory developed in the present paper may be summarized as follows: The thrust of the theory of the present paper lies in the development of a formalism, via the use of ring/scheme structures reconstructed via mono- anabelian geometry, in which the “dismantling/compressing nature” of the logarithm operation discussed above [cf. the Frobenius morphism in pos- itive characteristic] is “reorganized” in an abstract combinatorial fashion 4 SHINICHI MOCHIZUKI that exhibits the logarithm as a global operation on a number field which, moreover, is a sort of “isomomorphism up to an appropriate renormalization operation” [cf. the isomorphism induced on differen- tials by a Frobenius lifting, once one divides by p]. One important aspect of this theory is the analogy between this theory and [the positive characteristic portion of] p-adic Teichmüller theory [cf. §I5 below], in which the “naive pull-back” of an indigenous bundle by Frobenius never yields a bundle isomorphic to the original indigenous bundle, but the “renormalized Frobe- nius pull-back” does, in certain cases, allow one to obtain an output bundle that is isomorphic to the original input bundle. At a more detailed level, the main results of the present paper may be sum- marized as follows: In §1, we develop the absolute anabelian algorithms that will be necessary in our theory. In particular, we obtain a semi-absolute group-theoretic recon- struction algorithm [cf. Theorem 1.9, Corollary 1.10] for hyperbolic orbicurves of strictly Belyi type [cf. [Mzk21], Definition 3.5] over sub-p-adic fields i.e., such as number fields and nonarchimedean completions of number fields that is func- torial with respect to base-change of the base field. Moreover, we observe that the only “non-elementary” ingredient of these algorithms is the technique of Belyi cuspidalization developed in [Mzk21], §3, which depends on the main results of [Mzk5] [cf. Remark 1.11.3]. If one eliminates this non-elementary ingredient from these algorithms, then, in the case of function fields, one obtains a very elemen- tary semi-absolute group-theoretic reconstruction algorithm [cf. Theorem 1.11], which is valid over somewhat more general base fields, namely base fields which are “Kummer-faithful” [cf. Definition 1.5]. The results of §1 are of interest as anabelian results in their own right, independent of the theory of later portions of the present paper. For instance, it is hoped that elementary results such as Theorem 1.11 may be of use in introductions to anabelian geometry for advanced undergraduates or non-specialists [cf. [Mzk8], §1]. In §2, we develop an archimedean i.e., complex analytic analogue of the theory of §1. One important theme in this theory is the definition of “archimedean structures” which, like profinite Galois groups, are “immune to the ring structure- dismantling and compressing nature of the logarithm”. For instance, the notion that constitutes the archimedean counterpart to the notion of a profinite Galois group is the notion of an Aut-holomorphic structure [cf. Definition 2.1; Propo- sition 2.2; Corollary 2.3], which was motivated by the category-theoretic approach to holomorphic structures via the use of the topological group SL 2 (R) given in [Mzk14], §1. In this context, one central fact is the rather elementary observation that the group of holomorphic or anti-holomorphic automorphisms of the unit disc in the complex plane is commensurably terminal [cf. [Mzk20], §0] in the group of self-homeomorphisms of the unit disc [cf. Proposition 2.2, (ii)]. We also give an “al- gorithmic refinement” of the “parallelograms, rectangles, squares approach” of [Mzk14], §2 [cf. Propositions 2.5, 2.6]. By combining these two approaches and applying the technique of elliptic cuspidalization developed in [Mzk21], §3, we ob- tain a certain reconstruction algorithm [cf. Corollary 2.7] for the “local linear holomorphic structure” of an Aut-holomorphic orbispace arising from an el- liptically admissible [cf. [Mzk21], Definition 3.1] hyperbolic orbicurve, which is TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 5 compatible with the global portion of the Galois-theoretic theory of §1 [cf. Corollaries 2.8, 2.9]. In §3, §4, we develop the category-theoretic formalism centering around the notions of observables, telecores, and cores [cf. Definition 3.5] that are applied to express the compatibility of the “mono-anabelian” construction algo- rithms of §1 [cf. Corollary 3.6] and §2 [cf. Corollary 4.5] with the “log-Frobenius functor log” [in essence, a version of the usual “logarithm” at the various nonar- chimedean and archimedean primes of a number field]. We also study the failure of log-Frobenius compatibility that occurs if one attempts to take the “conventional anabelian” which we shall refer to as “bi-anabelian” approach to the situation [cf. Corollary 3.7]. Finally, in the remarks following Corollaries 3.6, 3.7, we discuss in detail the meaning of the various new category-theoretic notions that are intro- duced, as well as the various aspects of the analogy between these notions, in the context of Corollaries 3.6, 3.7, and the classical p-adic theory of the MF -objects of [Falt]. In §5, we develop a global formalism over number fields in which we study the canonical rigid integral structures i.e., log-shells that are obtained by applying the log-Frobenius compatibility discussed in §3, §4. These log-shells satisfy the following important properties: (L1) a log-shell is compact and hence of finite “log-volume” [cf. Corollary 5.10, (i)]; (L2) the log-volumes of (L1) are compatible with application of the log- Frobenius functor [cf. Corollary 5.10, (ii)]; (L3) log-shells are compatible with the operation of “panalocalization”, i.e., the operation of restricting to the disjoint union of the various primes of a number field in such a way that one “forgets” the global structure of the number field [cf. Corollary 5.5; Corollary 5.10, (iii)]; (L4) log-shells are compatible with the operation of “mono-analyticization”, i.e., the operation of “disabling the rigidity” of one of the “two combina- torial dimensions” of a ring, an operation that corresponds to allowing “Teichmüller dilations” in complex and p-adic Teichmüller theory [cf. Corollary 5.10, (iv)]. In particular, we note that property (L3) may be thought of as a rigidity property for certain global arithmetic line bundles [more precisely, the trivial arithmetic line bundle cf. Remarks 5.4.2, 5.4.3] that is analogous to the very strong i.e., by comparison to the behavior of arbitrary vector bundles on a curve rigidity prop- erties satisfied by MF -objects with respect to Zariski localization. Such rigidity properties may be thought of as a sort of “freezing of integral structures” with respect to Zariski localization [cf. Remark 5.10.2, (i)]. Finally, we discuss in some detail [cf. Remark 5.10.3] the analogy centering around the correspondence number field F once-punctured ell. curve X over F ←→ hyperbolic curve C in pos. char. ←→ nilp. ord. indig. bundle P over C between the theory of the present paper [involving hyperbolic orbicurves related to once-punctured elliptic curves over a number field] and the p-adic Teichmüller 6 SHINICHI MOCHIZUKI theory of [Mzk1], [Mzk4] [involving nilpotent ordinary indigenous bundles over hyperbolic curves in positive characteristic]. Finally, in an Appendix to the present paper, we expose the portion of the well-known theory of abelian varieties with complex multiplication [cf., e.g., [Lang- CM], [Milne-CM], for more details] that underlies the observation “(∗ CM )” related to the author by A. Tamagawa [cf. [Mzk20], Remark 3.8.1]. In particular, we verify that this observation (∗ CM ) does indeed hold. This implies that the observation “(∗ A-qLT )” discussed in [Mzk20], Remark 3.8.1, also holds, and hence, in particular, that the hypothesis of [Mzk20], Corollary 3.9, to the effect that “either (∗ A-qLT ) or (∗ CM ) holds” may be eliminated [i.e., that [Mzk20], Corollary 3.9, holds uncondi- tionally]. Although the content of this Appendix is not directly technically related to the remainder of the present paper, the global arithmetic nature of the content of this Appendix, as well as the accompanying discussion of the relationship of this global content with considerations in p-adic Hodge theory, is closely related in spirit to the analogies between the content of the remainder of the present paper and the theory of earlier papers in the present series of papers, i.e., more precisely, [Mzk20], §3; [Mzk21], §2. §I2. Fundamental Naive Questions Concerning Anabelian Geometry One interesting aspect of the theory of the present paper is that it is intimately related to various fundamental questions concerning anabelian geometry that are frequently posed by newcomers to the subject. Typical examples of these fundamen- tal questions are the following: (Q1) Why is it useful or meaningful to study anabelian geometry in the first place? (Q2) What exactly is meant by the term “group-theoretic reconstruction” in discussions of anabelian geometry? (Q3) What is the significance of studying anabelian geometry over mixed- characteristic local fields [i.e., p-adic local fields] as opposed to num- ber fields? (Q4) Why is birational anabelian geometry insufficient i.e., what is the significance of studying the anabelian geometry of hyperbolic curves, as opposed to their function fields? In fact, the answers to these questions that are furnished by the theory of the present paper are closely related. As was discussed in §I1, the answer to (Q1), from the point of view of the present paper, is that anabelian geometry more specifically, “mono-anabelian geometry” provides a framework that is sufficiently well-endowed as to contain “data reminiscent of the data constituted by various scheme-theoretic structures”, but has the virtue of being based not on ring structures, but rather on profinite [Galois] groups, which are “neutral” with respect to the operation of taking the logarithm. TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 7 The answer to (Q2) is related to the algorithmic approach to absolute anabelian geometry taken in the present three-part series [cf. the Introduction to [Mzk20]]. That is to say, typically, in discussions concerning “Grothendieck Conjecture-type fully faithfulness results” [cf., e.g., [Mzk5]] the term “group-theoretic reconstruc- tion” is defined simply to mean “preserved by an arbitrary isomorphism between the étale fundamental groups of the two schemes under consideration”. This point of view will be referred to in the present paper as “bi-anabelian”. By contrast, the algorithmic approach to absolute anabelian geometry involves the development of “software” whose input data consists solely of, for instance, a single abstract profinite group [that just happens to be isomorphic to the étale fundamental group of a scheme], and whose output data consists of various structures reminiscent of scheme theory [cf. the Introduction to [Mzk20]]. This point of view will be referred to in the present paper as “mono-anabelian”. Here, the mono-anabelian “soft- ware” is required to be functorial, e.g., with respect to isomorphisms of profinite groups. Thus, it follows formally that “mono-anabelian” =⇒ “bi-anabelian” [cf. Remark 1.9.8]. On the other hand, although it is difficult to formulate such issues completely precisely, the theory of the present paper [cf., especially, §3] sug- gests strongly that the opposite implication should be regarded as false. That is to say, whereas the mono-anabelian approach yields a framework that is “neutral” with respect to the operation of taking the logarithm, the bi-anabelian approach fails to yield such a framework [cf. Corollaries 3.6, 3.7, and the following remarks; §I4 below]. Here, we pause to remark that, in fact, although, historically speaking, many theorems were originally formulated in a “bi-anabelian” fashion, careful inspec- tion of their proofs typically leads to the recovery of “mono-anabelian algorithms”. Nevertheless, since formulating theorems in a “mono-anabelian” fashion, as we have attempted to do in the present paper [and more generally in the present three-part series, but cf. the final portion of the Introduction to [Mzk21]], can be quite cumber- some and indeed is one of the main reasons for the unfortunately lengthy nature of the present paper! it is often convenient to formulate final theorems in a “bi- anabelian” fashion. On the other hand, we note that the famous Neukirch-Uchida theorem on the anabelian nature of number fields appears to be one important counterexample to the above remark. That is to say, to the author’s knowledge, proofs of this result never yield “explicit mono-anabelian reconstruction algorithms of the given number field”; by contrast, Theorem 1.9 of the present paper does give such an explicit construction of the “given number field” [cf. Remark 1.9.5]. Another interesting aspect of the algorithmic approach to anabelian geometry is that one may think of the “software” constituted by such algorithms as a sort of “combinatorialization” of the original schemes [cf. Remark 1.9.7]. This point of view is reminiscent of the operation of passing from a “scheme-theoretic” MF - object to an associated Galois representation, as well as the general theme in various papers of the author concerning a “category-theoretic approach to scheme theory” [cf., e.g., [Mzk13], [Mzk14], [Mzk16], [Mzk17], [Mzk18]] of “extracting from scheme- theoretic arithmetic geometry the abstract combinatorial patterns that underlie the scheme theory”. 8 SHINICHI MOCHIZUKI The answer to (Q3) provided by the theory of the present paper is that the absolute p-adic [mono-]anabelian results of §1 underlie the panalocalizability of log-shells discussed in §I1 [cf. property (L3)]. Put another way, these results imply that the “geometric framework immune to the application of the logarithm” i.e., immune to the dismantling of the “” and “” dimensions of a ring discussed in §I1 may be applied locally at each prime of a number field regarded as an isolated entity, i.e., without making use of the global structure of the number field cf. the discussion of “freezing of integral structures” with respect to Zariski localization in Remark 5.10.2, (i). For more on the significance of the operation of passing “  ” in the context of nonarchimedean log-shells i.e., the operation of passing “O k ×  log k (O k × )” we refer to the discussion of nonarchimedean log-shells in §I3 below. The answer to (Q4) furnished by the theory of the present paper [cf. Remark 1.11.4] i.e., one fundamental difference between birational anabelian geometry and the anabelian geometry of hyperbolic curves is that [unlike spectra of func- tion fields!] “most” hyperbolic curves admit “cores” [in the sense of [Mzk3], §3; [Mzk10], §2], which may be thought of as a sort of abstract “covering-theoretic” analogue [cf. Remark 1.11.4, (ii)] of the notion of a “canonical rigid integral struc- ture” [cf. the discussion of log-shells in §I1]. Moreover, if one attempts to work with the Galois group of a function field supplemented by some additional structure such as the set of cusps arising from scheme theory! that determines a hyperbolic curve structure, then one must sacrifice the crucial mono-anabelian nature of one’s reconstruction algorithms [cf. Remarks 1.11.5; 3.7.7, (ii)]. Finally, we observe that there certainly exist many “fundamental naive ques- tions” concerning anabelian geometry for which the theory of the present paper does not furnish any answers. Typical examples of such fundamental questions are the following: (Q5) What is the significance of studying the anabelian geometry of proper hyperbolic curves, as opposed to affine hyperbolic curves? (Q6) What is the significance of studying pro-Σ [where Σ is some nonempty set of prime numbers] anabelian geometry, as opposed to profinite anabelian geometry [cf., e.g., Remark 3.7.6 for a discussion of why pro-Σ anabelian geometry is ill-suited to the needs of the theory of the present paper]? (Q7) What is the significance of studying anabelian geometry in positive characteristic, e.g., over finite fields? It would certainly be of interest if further developments could shed light on these questions. §I3. Dismantling the Two Combinatorial Dimensions of a Ring As was discussed in §I1, a ring may be thought of as a mathematical object that consists of “two combinatorial dimensions”, corresponding to its additive structure  and its multiplicative structure  [cf. Remark 5.6.1, (i)]. When the ring un- der consideration is a [say, for simplicity, totally imaginary] number field F or TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 9 a mixed-characteristic nonarchimedean local field k, these two combinatorial dimensions may also be thought of as corresponding to the two cohomological dimensions of the absolute Galois groups G F , G k of F , k [cf. [NSW], Proposition 8.3.17; [NSW], Theorem 7.1.8, (i)]. In a similar vein, when the ring under con- sideration is a complex archimedean field k ( = C), then the two combinatorial dimensions of k may also be thought of as corresponding to the two topological i.e., real dimensions of the underlying topological space of the topological group k × . Note that in the case where the local field k is nonarchimedean (respec- tively, archimedean), precisely one of the two cohomological (respectively, real) dimensions of G k (respectively, k × ) namely, the dimension corresponding to the  · Fr [generated by the Frobenius element] (re- maximal unramified quotient G k  Z spectively, the topological subgroup of units S 1 = O k × k × ) is rigid with respect to, say, automorphisms of the topological group G k (respectively, k × ), while the other dimension namely, the dimension corresponding to the inertia subgroup I k G k (respectively, the value group k ×  R >0 ) is not rigid [cf. Remark 1.9.4]. [In the nonarchimedean case, this phenomenon is discussed in more detail in [NSW], the Closing Remark preceding Theorem 12.2.7.] Thus, each of the various nonarchimedean “G k ’s” and archimedean “k × ’s” that arise at the various primes of a number field may be thought of as being a sort of “arithmetic G m i.e., an abstract arithmetic “cylinder” that decomposes into a [twisted] product of  · Fr, k ×  R >0 ] “units” [i.e., I k G k , O k × k × ] and value group [i.e., G k  Z ‘arithmetic G m     ‘units’ ‘×’  ‘×’ ‘value group’  with the property that one of these two factors is rigid, while the other is not. Here, it is interesting to note that the correspondence between units/value group and rigid/non-rigid differs [i.e., “goes in the opposite direction”] in the nonarchimedean and archimedean cases. This phenomenon is reminiscent of the product formula in elementary number theory, as well as of the behavior of the log-Frobenius functor log at nonarchimedean versus archimedean primes [cf. Remark 4.5.2; the discussion of log-shells in the final portion of the present §I3]. C × G k rigid S 1 × non-rigid I k non-rigid R >0  rigid  · Fr Z On the other hand, the perfection of the topological group obtained as the image of the non-rigid portion I k in the abelianization G ab k of G k is naturally iso- morphic, by local class field theory, to k. Moreover, by the theory of [Mzk2], the 10 SHINICHI MOCHIZUKI decomposition of this copy of k [i.e., into sets of elements with some given p-adic val- uation] determined by the p-adic valuation on k may be thought of as corresponding to the ramification filtration on G k and is precisely the data that is “deformed” by automorphisms of k that do not arise from field automorphisms. That is to say, this aspect of the non-rigidity of G k is quite reminiscent of the non-rigidity of the topological group R >0 [i.e., of the non-rigidity of the structure on this topological group arising from the usual archimedean valuation on R, which determines an isomorphism between this topological group and some “fixed, standard copy” of R >0 ]. In this context, one of the first important points of the “mono-anabelian the- ory” of §1, §2 of the present paper is that if one supplements a(n) nonarchimedean G k (respectively, archimedean k × ) with the data arising from a hyperbolic or- bicurve [which satisfies certain properties cf. Corollaries 1.10, 2.7], then this supplementary data has the effect of rigidifying both dimensions of G k (re- spectively, k × ). In the case of [a nonarchimedean] G k , this data consists of the outer action of G k on the profinite geometric fundamental group of the hyperbolic orbicurve; in the case of [an archimedean] k × , this data consists, in essence, of the various local actions of open neighborhoods of the origin of k × on the squares or rectangles [that lie in the underlying topological [orbi]space of the Riemann [orbi]surface determined by the hyperbolic [orbi]curve] that encode the holomor- phic structure of the Riemann [orbi]surface [cf. the theory of [Mzk14], §2]. Here, it is interesting to note that these “rigidifying actions” are reminiscent of the discussion of “hidden endomorphisms” in the Introduction to [Mzk21], as well as of the discussion of “intrinsic Hodge theory” in the context of p-adic Teichmüller theory in [Mzk4], §0.10.       C × G k rigidify ‘’ geometric data representing hyp. curve  rectangles/squares on hyp. Riemann surface out profinite geometric fund. gp. of hyp. curve  Thus, in summary, the “rigidifying actions” discussed above may be thought of as constituting a sort of “arithmetic holomorphic structure” on a nonar- chimedean G k or an archimedean k × . This arithmetic holomorphic structure is immune to the log-Frobenius operation log [cf. the discussion of §I1], i.e., immune to the “juggling of , ” effected by log [cf. the illustration of Remark 5.10.2, (iii)]. TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 11 On the other hand, if one exits such a “zone of arithmetic holomorphy” an operation that we shall refer to as mono-analyticization then a nonar- chimedean G k or an archimedean k × is stripped of the rigidity imparted by the above rigidifying actions, hence may be thought of as being subject to Teichmüller di- lations [cf. Remark 5.10.2, (ii), (iii)]. Indeed, this is intuitively evident in the archimedean case, in which the quotient k ×  R >0 is subject [i.e., upon mono- analyticization, so k × is only considered as a topological group] to automorphisms x → x λ R >0 , for λ R >0 . If, moreover, one thinks of of the form R >0 the value groups of archimedean and nonarchimedean primes as being “synchro- nized” [so as to keep from violating the product formula which plays a crucial role in the theory of “heights”, i.e., degrees of global arithmetic line bundles], then the operation of mono-analyticization necessarily results in analogous “Teichmüller dilations” at nonarchimedean primes. In the context of the theory of Frobenioids, such Teichmüller dilations [whether archimedean or nonarchimedean] correspond to the unit-linear Frobenius functor studied in [Mzk16], Proposition 2.5. Note that the “non-linear juggling of ,  by log within a zone of arithmetic holomorphy” and the “linear Teichmüller dilations inherent in the operation of mono-analyticization” are reminiscent of the Riemannian geometry of the upper half-plane, i.e., if one thinks of “juggling” as corresponding to rotations at a point, and “dilations” as corre- sponding to geodesic flows originating from the point.  log   mono-analyticization [linear] Teichmüller dilation --------------------------------------- Put another way, the operation of mono-analyticization may be thought of as an operation on the “arithmetic holomorphic structures” discussed above that forms a sort of arithmetic analogue of the operation of passing to the underlying real analytic manifold of a Riemann surface. number fields and their localizations Riemann surfaces “arithmetic holomorphic structures” via rigidifying hyp. curves the operation of mono-analyticization complex holomorphic structure on the Riemann surface passing to the underlying real analytic manifold Thus, from this point of view, one may think of the disjoint union of the various G k ’s, k × ’s over the various nonarchimdean and archimedean primes of the number field as being the “arithmetic underlying real analytic manifold” of the “arith- metic Riemann surface” constituted by the number field. Indeed, it is precisely this sort of disjoint union that arises in the theory of mono-analyticization, as developed in §5. 12 SHINICHI MOCHIZUKI Next, we consider the effect on log-shells of the operation of mono-analyticization. In the nonarchimedean case, log k (O k × ) = O k × /(torsion) may be reconstructed group-theoretically from G k as the quotient by torsion of the image of I k in the abelianization G ab k [cf. Proposition 5.8, (i), (ii)]; a similar con- struction may be applied to finite extensions k of k. Moreover, this construction involves only the group of units O k × [i.e., it does not involve the value groups, which, as discussed above, are subject to Teichmüller dilations], hence is compatible with the operation of mono-analyticization. Thus, this construction yields a canonical rigid integral structure, i.e., in the form of the topological module log k (O k × ), which may be thought of as a sort of approximation of some nonarchimedean localization of the trivial global arithmetic line bundle [cf. Remarks 5.4.2, 5.4.3] that is achieved without the use of [the two combinatorial dimensions of] the ring structure on O k . Note, moreover, that the ring structure on the perfec- tion log k (O k × ) pf [i.e., in effect, “log k (O k × ) Q”] of this module is obliterated by the operation of mono-analyticization. That is to say, this ring structure is only acces- sible within a “zone of arithmetic holomorphy” [as discussed above]. On the other hand, if one returns to such a zone of arithmetic holomorphy to avail oneself of the ring structure on log k (O k × ) pf , then applying the operation of mono-analyticization amounts to applying the construction discussed above to the group of units of log k (O k × ) pf [equipped with the ring structure furnished by the zone of arithmetic holomorphy under consideration]. That is to say, the freedom to execute, at will, both the operations of exiting and re-entering zones of arithmetic holomorphy is inextricably linked to the “juggling of , ” via log [cf. Remark 5.10.2, (ii), (iii)]. In the archimedean case, if one writes log(O k × )  O k × for the universal covering topological group of O k × [i.e., in essence, the exponential map “2πi · R  S 1 ”], then the surjection log(O k × )  O k × determines on log(O k × ) a “canonical rigid line segment of length 2π”. Thus, if one writes k = k im × k rl for the product decomposition of the additive topological group k into imaginary [i.e., “i · R”] and real [ı.e., “R”] parts, then we obtain a natural isometry log(O k × ) × log(O k × ) k im × k rl = k [i.e., the product of the identity isomorphism 2πi · R = i · R and the isomorphism 2πi · R R given by dividing by ±i] which is well-defined up to multiplication by ±1 on the second factors [cf. Definition 5.6, (iv); Proposition 5.8, (iv), (v)]. In particular, “log(O k × ) × log(O k × )” may be regarded as a construction, based on the “rigid” topological group O k × [which is not subject to Teichmüller dilations!], of a canonical rigid integral structure [determined by the canonical rigid line segments discussed above] that serves as an approximation of some archimedean localization of the trivial global arithmetic line bundle and, moreover, is compatible with the operation of mono-analyticization [cf. the nonarchimedean case]. On the TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 13 other hand, [as might be expected by comparison to the nonarchimedean case] once one exits a zone of arithmetic holomorphy, the ±1-indeterminacy that occurs in the above natural isometry has the effect of obstructing any attempts to transport the ring structure of k via this natural isometry so as to obtain a structure of complex archimedean field on log(O k × ) × log(O k × ) [cf. Remark 5.8.1]. Finally, just as in the nonarchimedean case, the freedom to execute, at will, both the operations of exiting and re-entering zones of arithmetic holomorphy is inextricably linked to the “juggling of , ” via log [cf. Remark 5.10.2, (ii), (iii)] a phenomenon that is strongly reminiscent of the crucial role played by rotations in the theory of mono-analyticizations of archimedean log-shells [cf. Remark 5.8.1]. §I4. Mono-anabelian Log-Frobenius Compatibility Within each zone of arithmetic holomorphy, one wishes to apply the log- Frobenius functor log. As discussed in §I1, log may be thought of as a sort of “wall” that may be penetrated by such “elementary combinatorial/topological ob- jects” as Galois groups [in the nonarchimedean case] or underlying topological spaces [in the archimedean case], but not by rings or functions [cf. Remark 3.7.7]. This situation suggests a possible analogy with ideas from physics in which “étale-like” structures [cf. the Introduction of [Mzk16]], which can penetrate the log-wall, are regarded as “massless”, like light, while “Frobenius-like” structures [cf. the Intro- duction of [Mzk16]], which cannot penetrate the log-wall, are regarded as being of “positive mass”, like ordinary matter [cf. Remark 3.7.5, (iii)]. Galois groups, = topological spaces ------------------- log log Galois groups, = topological spaces ------------------- log rings, functions  log  rings, functions In the archimedean case, since topological spaces alone are not sufficient to trans- port “holomorphic structures” in the usual sense, we take the approach in §2 of con- sidering “Aut-holomorphic spaces”, i.e., underlying topological spaces of Riemann surfaces equipped with the additional data of a group of “special self-homeomor- phisms” [i.e., bi-holomorphic automorphisms] of each [sufficiently small] open con- nected subset [cf. Definition 2.1, (i)]. The point here is to “somehow encode the usual notion of a holomorphic structure” in such a way that one does not need to resort to the use of “fixed reference models” of the field of complex numbers C [as is done in the conventional definition of a holomorphic structure, which consists of local comparison to such a fixed reference model of C], since such models of C fail to be “immune” to the application of log cf. Remarks 2.1.2, 2.7.4. This situation is very much an archimedean analogue of the distinction between mono- anabelian and bi-anabelian geometry. That is to say, if one thinks of one of the 14 SHINICHI MOCHIZUKI two schemes that occur in bi-anabelian comparison results as the “given scheme of interest” and the other scheme as a “fixed reference model”, then although these two schemes are related to one another via purely Galois-theoretic data, the scheme structure of the “scheme of interest” is reconstructed from the Galois-theoretic data by transporting the scheme structure of “model scheme”, hence requires the use of input data [i.e., the scheme structure of the “model scheme”] that cannot penetrate the log-wall. In order to formalize these ideas concerning the issue of distinguishing between “model-dependent”, “bi-anabelian” approaches and “model-independent”, “mono- anabelian” approaches, we take the point of view, in §3, §4, of considering “series of operations” in the form of diagrams [parametrized by various oriented graphs] of functors applied to various “types of data” in the form of objects of categories [cf. Remark 3.6.7]. Although, by definition, it is impossible to compare the “different types of data” obtained by applying these various “operations”, if one considers “projections” of these operations between different types of data onto morphisms between objects of a single category [i.e., a single “type of data”], then such comparisons become possible. Such a “projection” is formalized in Definition 3.5, (iii), as the notion of an observable. One special type of observable that is of crucial importance in the theory of the present paper is an observable that “captures a certain portion of various distinct types of data that remains constant, up to isomorphism, throughout the series of operations applied to these distinct types of data”. Such an observable is referred to as a core [cf. Definition 3.5, (iii)]. Another important notion in the theory of the present paper is the notion of telecore [cf. Definition 3.5, (iv)], which may be thought of as a sort of “core structure whose compatibility apparatus [i.e., ‘constant nature’] only goes into effect after a certain time lag” [cf. Remark 3.5.1]. Before explaining how these notions are applied in the situation over number fields considered in the present paper, it is useful to consider the analogy between these notions and the classical p-adic theory. The prototype of the notion of a core is the constant nature [i.e., up to equivalence of categories] of the étale site of a scheme in positive char- acteristic with respect to the [operation constituted by the] Frobenius morphism. Put another way, cores may be thought of as corresponding to the notion of “slope zero” Galois representations in the p-adic theory. By contrast, telecores may be thought of as corresponding to the notion of “positive slope” in the p-adic theory. In particular, the “time lag” inherent in the compatibility apparatus of a telecore may be thought of as corresponding to the “lag”, in terms of powers of p, that occurs when one applies Hensel’s lemma [cf., e.g., [Mzk21], Lemma 2.1] to lift solutions, modulo various powers of p, of a polynomial equation that gives rise to a crystalline Galois representation e.g., arising from an “MF -object” of [Falt] for which the slopes of the Frobenius action are positive [cf. Remark 3.6.5 for more on this topic]. This formal analogy with the classical p-adic theory forms the starting point for the analogy with p-adic Teichmüller theory to be discussed in §I5 below [cf. Remark 3.7.2]. TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 15 Now let us return to the situation involving k, k, G k , and log k discussed at the beginning of §I1. Suppose further that we are given a hyperbolic orbicurve over k as in the discussion of §I3, whose étale fundamental group Π surjects onto G k × [hence may be regarded as acting on k, k ] and, moreover, satisfies the important property of rigidifying G k [as discussed in §I3]. Then the “series of operations” performed in this context may be summarized as follows [cf. Remark 3.7.3, (ii)]: Π    Π  Π × × k An k    Π Π   × k An log Here, the various operations “”, “” may be described in words as follows: (O1) One applies the mono-anabelian reconstruction algorithms of §1 × × × to Π to construct a “mono-anabelian copy” k An of k . Here, k An is the group of nonzero elements of a field k An . Moreover, it is important to note that k An is equipped with the structure not of “some field k An isomorphic to k”, but rather of “the specific field [isomorphic to k] reconstructed via the mono-anabelian reconstruction algorithms of §1”. (O2) One forgets the fact that k An arises from the mono-anabelian recon- struction algorithms of §1, i.e., one regards k An just as “some field k  [isomorphic to k]”. (O3) Having performed the operation of (O2), one can now proceed to apply log-Frobenius operation log [i.e., log k ] to k  . This operation log may be thought of as the assignment ×  k  )   {log k  (O × ) pf } × ) k  that maps the group of nonzero elements of the topological field k  to the group of nonzero elements of the topological field “log k  (O × ) pf [cf. the k  discussion of §I3]. (O4) One forgets all the data except for the profinite group Π. (O5) This is the same operation as the operation described in (O1). With regard to the operation log, observe that if we forget the various field or group structures involved, then the arrows × k  ← O × k  log k  (O × ) pf k  ← {log k  (O × ) pf } × k  allow one to relate the input of log [on the left] to the the output of log [on the right]. That is to say, in the formalism developed in §3, these arrows may be regarded as defining an observable “S log associated to log [cf. Corollary 3.6, (iii)]. 16 SHINICHI MOCHIZUKI If one allows oneself to reiterate the operation log, then one obtains diagrams equipped with a natural Z-action [cf. Corollary 3.6, (v)]. These diagrams equipped with a Z-action are reminiscent, at a combinatorial level, of the “arithmetic G m ’s” that occurred in the discussion of §I3 [cf. Remark 3.6.3]. Next, observe that the operation of “projecting to Π” [i.e., forgetting all of the data under consideration except for Π] is compatible with the execution of any of these operations (O1), (O2), (O3), (O4), (O5). That is to say, Π determines a core of this collection of operations [cf. Corollary 3.6, (i), (ii), (iii)]. Moreover, since the mono-anabelian reconstruction algorithms of §1 are “purely group-theoretic” and depend only on the input data constituted by Π, it follows immediately that [by × “projecting to Π” and then applying these algorithms] “(Π  k An )” also forms a core of this collection of operations [cf. Corollary 3.6, (i), (ii), (iii)]. In particular, × we obtain a natural isomorphism between the “(Π  k An )’s” that occur following the first and fourth “’s” of the above diagram. On the other hand, the “forgetting” operation of (O2) may be thought of as a × sort of section of the “projection to the core  k An )”. This sort of section will be referred to as a telecore; a telecore frequently comes equipped with an auxiliary structure, called a contact structure, which corresponds in the present situation to the isomorphism of underlying fields [stripped of their respective zero elements] × × × k An k  [cf. Corollary 3.6, (ii)]. Even though the core “(Π  k An )”, regarded as an object obtained by projecting, is constant [up to isomorphism], the section obtained in this way does not yield a “constant” collection of data [with respect to the operations of the diagram above] that is compatible with the observable S log . × Indeed, forgetting the marker “An” of [the constant] k An and then applying log is not compatible, relative to S log , with forgetting the marker “An” i.e., since log obliterates the ring structures involved [cf. Corollary 3.6, (iv); Remark 3.6.1]. Nevertheless, if, subsequent to applying the operations of (O2), (O3), one × projects back down to “(Π  k An )”, then, as was observed above, one obtains a × natural isomorphism between the initial and final copies of “(Π  k An )”. It is in this sense that one may think of a telecore as a “core with a time lag”. One way to summarize the above discussion is as follows: The “purely group- theoretic” mono-anabelian reconstruction algorithms of §1 allow one to construct × models of scheme-theoretic data [i.e., the “k An ”] that satisfy the following three properties [cf. Remark 3.7.3, (i), (ii)]: × (P1) coricity [i.e., the “property of being a core” of “Π”, “(Π  k An )”]; (P2) comparability [i.e., via the telecore and contact structures discussed × above] with log-subject copies [i.e., the “k  ”, which are subject to the action of log]; (P3) log-observability [i.e., via “S log ”]. One way to understand better what is gained by this mono-anabelian approach is to consider what happens if one takes a bi-anabelian approach to this situation [cf. Remarks 3.7.3, (iii), (iv); 3.7.5]. TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 17 In the bi-anabelian approach, instead of taking just “Π” as one’s core, one takes the data ×  k model ) where “k model is some fixed reference model of k as one’s core [cf. Corollary 3.7, (i)]. The bi-anabelian version [i.e., fully faithfulness in the style of the “Grothendieck Conjecture”] of the mono-anabelian theory of §1 then gives rise × × to telecore and contact structures by considering the isomorphism k model k  × × arising from an isomorphism between the “Π’s” that act on k model , k  [cf. Corollary 3.7, (ii)]. Moreover, one may define an observable “S log as in the mono-anabelian case [cf. Corollary 3.7, (iii)]. Just as in the mono-anabelian case, since log obliterates × the ring structures involved, this model k model fails to be simultaneously compatible with the observable S log and the telecore and [a slight extension, as described in Corollary 3.7, (ii), of the] contact structures just mentioned [cf. Corollary 3.7, (iv)]. On the other hand, whereas in the mono-anabelian case, one may recover from this failure of compatibility by projecting back down to “Π” [which remains × × intact!] and hence to “(Π  k An )”, in the bi-anabelian case, the “k model portion × of “the core  k model )” which is an essential portion of the input data for reconstruction algorithms via the bi-anabelian approach! [cf. Remarks 3.7.3, (iv); 3.7.5, (ii)] is obliterated by log, thus rendering it impossible to relate the × “(Π  k model )’s” before and after the application of log via an isomorphism that is compatible with all of the operations involved. At a more technical level, the non- existence of such a natural isomorphism may be seen in the fact that the coricity of × “(Π  k model )” is only asserted in Corollary 3.7, (i), for a certain limited portion of the diagram involving “all of the operations under consideration” [cf. also the incompatibilities of Corollary 3.7, (iv)]. This contrasts with the [manifest!] coricity × of “Π”, “(Π  k An )” with respect to all of the operations under consideration in the mono-anabelian case [cf. Corollary 3.6, (i), (ii), (iii)]. In this context, one important observation is that if one tries to “subsume” the × model “k model into Π by “regarding” this model as an object that “arises from the sole input data Π”, then one must contend with various problems from the point of view of functoriality cf. Remark 3.7.4 for more details on such “functorially × trivial models”. That is to say, to regard “k model in this way means that one must contend with a situation in which the functorially induced action of Π on × “k model is trivial! Finally, we note in passing that the “dynamics” of the various diagrams of operations [i.e., functors] appearing in the above discussion are reminiscent of the analogy with physics discussed at the beginning of the present §I4 i.e., that “Π” × is massless, like light, while “k  is of positive mass. §I5. Analogy with p-adic Teichmüller Theory We have already discussed in §I1 the analogy between the log-Frobenius oper- ation log and the Frobenius morphism in positive characteristic. This analogy may be developed further [cf. Remarks 3.6.6, 3.7.2 for more details] into an analogy be- tween the formalism discussed in §I4 and the notion of a uniformizing MF -object 18 SHINICHI MOCHIZUKI as discussed in [Mzk1], [Mzk4], i.e., an MF -object in the sense of [Falt] that gives rise to “canonical coordinates” that may be regarded as a sort of p-adic uni- formization of the variety under consideration. Indeed, in the notation of §I4, the × “mono-anabelian output data  k An )” may be regarded as corresponding to the “Galois representation” associated to a structure of “uniformizing MF - × object” on the scheme-theoretic “(Π  k  )”. The telecore structures discussed in §I4 may be regarded as corresponding to a sort of Hodge filtration, i.e., an opera- tion relating the “Frobenius crystal” under consideration to a specific scheme theory × “(Π  k  )”, among the various scheme theories separated from one another by [the non-ring-homomorphism!] log. The associated contact structures then take on an appearance that is formally reminiscent of the notion of a connection in the classical crystalline theory. The failure of the log-observable, telecore, and contact structures to be simultaneously compatible [cf. Corollaries 3.6, (iv); 3.7, (iv)] may then be regarded as corresponding to the fact that, for instance in the case of the uniformizing MF -objects determined by indigenous bundles in [Mzk1], [Mzk4], the Kodaira-Spencer morphism is an isomorphism [i.e., the fact that the Hodge filtration fails to be a horizontal Frobenius-invariant!]. mono-anabelian theory p-adic theory log-Frobenius log mono-anabelian output data telecore structure contact structure simultaneous incompatibility of log-observable, telecore, and contact structures Frobenius Frobenius-invariants Hodge filtration connection Kodaira-Spencer morphism of an indigenous bundle is an isomorphism In the context of this analogy, we observe that the failure of the logarithms at the various localizations of a number field to extend to a global map involving the number field [cf. Remark 5.4.1] may be regarded as corresponding to the fail- ure of various Frobenius liftings on affine opens [i.e., localizations] of a hyperbolic curve [over, say, a ring of Witt vectors of a perfect field] to extend to a morphism defined [“globally”] on the entire curve [cf. [Mzk21], Remark 2.6.2]. This lack of a global extension in the p-adic case means, in particular, that it does not make sense to pull-back arbitrary coherent sheaves on the curve via such Frobenius lift- ings. On the other hand, if a coherent sheaf on the curve is equipped with the structure of a crystal, then a “global pull-back of the crystal” is well-defined and “canonical”, even though the various local Frobenius liftings used to construct it are not. In a similar way, although the logarithms at localizations of a number field are not compatible with the ring structures involved, hence cannot be used to pull-back arbitrary ring/scheme-theoretic objects, they can be used to “pull-back” Galois-theoretic structures, such as those obtained by applying mono-anabelian re- construction algorithms. TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III mono-anabelian theory p-adic theory logarithms at localizations of a number field nonexistence of global logarithm on a number field incompatibility of log with ring structures compatibility of log with Galois, mono-anabelian algorithms the result of forgetting “An” [cf. (O2) of §I3] Frobenius liftings on affine opens of a hyperbolic curve nonexistence of global Frobenius lifting on a hyperbolic curve noncanonicality of local liftings of positive characteristic Frobenius Frobenius pull-back of crystals the underlying coherent sheaf of a crystal 19 Moreover, this analogy may be developed even further by specializing from arbitrary uniformizing MF -objects to the indigenous bundles of the p-adic Te- ichmüller theory of [Mzk1], [Mzk4]. To see this, we begin by observing that the non-rigid dimension of the localizations of a number field “G k ”, “k × in the discus- sion of §I3 may be regarded as analogous to the non-rigidity of a p-adic deformation of an affine open [i.e., a localization] of a hyperbolic curve in positive character- istic. If, on the other hand, such a hyperbolic curve is equipped with the crystal determined by a p-adic indigenous bundle, then, even if one restricts to an affine open, this filtered crystal has the effect of rigidifying a specific p-adic deformation of this affine open. Indeed, this rigidifying effect is an immediate consequence of the fact that the Kodaira-Spencer morphism of an indigenous bundle is an iso- morphism. Put another way, this Kodaira-Spencer isomorphism has the effect of allowing the affine open to “entrust its moduli” to the crystal determined by the p-adic indigenous bundle. This situation is reminiscent of the rigidifying ac- tions discussed in §I3 of “G k ”, “k × on certain geometric data arising from a hyperbolic orbicurve that is related to a once-punctured elliptic curve. That is to say, the mono-anabelian theory of §1, §2 allows these localizations “G k ”, “k × of a number field to “entrust their ring structures” i.e., their “arithmetic holomorphic moduli” to the hyperbolic orbicurve under consideration. This leads naturally [cf. Remark 5.10.3, (i)] to the analogy already referred to in §I1: mono-anabelian theory p-adic theory number field F once-punctured ell. curve X over F hyperbolic curve C in pos. char. nilp. ord. indig. bundle P over C If, moreover, one modifies the canonical rigid integral structures furnished by log- shells by means of the “Gaussian zeroes” [i.e., the inverse of the “Gaussian poles”] that appear in the Hodge-Arakelov theory of elliptic curves [cf., e.g., [Mzk6], §1.1], then one may further refine the above analogy by regarding indigenous bun- dles as corresponding to the crystalline theta object [which may be thought of as 20 SHINICHI MOCHIZUKI an object obtained by equipping a direct sum of trivial line bundles with the inte- gral structures determined by the Gaussian zeroes] of Hodge-Arakelov theory [cf. Remark 5.10.3, (ii)]. From this point of view, the mono-anabelian theory of §1, §2, which may be thought of as centering around the technique of Belyi cuspidal- izations, may be regarded as corresponding to the theory of indigenous bundles in positive characteristic [cf. [Mzk1], Chapter II], which centers around the Ver- schiebung on indigenous bundles. Moreover, the theory of the étale theta function given in [Mzk18], which centers around the technique of elliptic cuspidalizations, may be regarded as corresponding to the theory of the Frobenius action on square differentials in [Mzk1], Chapter II. Indeed, just as the technique of elliptic cuspidal- izations may be thought of a sort of linearized, simplified version of the technique of Belyi cuspidalizations, the Frobenius action on square differentials occurs as the derivative [i.e., a “linearized, simplified version”] of the Verschiebung on indige- nous bundles. For more on this analogy, we refer to Remark 5.10.3. In passing, we observe, relative to the point of view that the theory of the étale theta func- tion given in [Mzk18] somehow represents a “linearized, simplified version” of the mono-anabelian theory of the present paper, that the issue of mono- versus bi- anabelian geometry discussed in the present paper is vaguely reminiscent of the issue of mono- versus bi-theta environments, which constitutes a central theme in [Mzk18]. In this context, it is perhaps natural to regard the “log-wall” discussed in §I4 which forms the principal obstruction to applying the bi-anabelian approach in the present paper as corresponding to the “Θ-wall” constituted by the theta function between the theta and algebraic trivializations of a certain ample line bun- dle which forms the principal obstruction to the use of bi-theta environments in the theory of [Mzk18]. mono-anabelian theory p-adic theory crystalline theta objects in scheme-theoretic Hodge-Arakelov theory Belyi cuspidalizations in mono-anabelian theory of §1 elliptic cuspidalizations in the theory of the étale theta function [cf. [Mzk18]] scheme-theoretic indigenous bundles [cf. [Mzk1], Chapter I] Verschiebung on pos. char. indigenous bundles [cf. [Mzk1], Chapter II] Frobenius action on square differentials [cf. [Mzk1], Chapter II] Thus, in summary, the analogy discussed above may be regarded as an anal- ogy between the theory of the present paper and the positive characteristic portion of the theory of [Mzk1]. This “positive characteristic portion” may be regarded as including, in a certain sense, the “liftings modulo p 2 portion” of the theory of [Mzk1] since this “liftings modulo p 2 portion” may be formulated, to a certain extent, in terms of positive characteristic scheme theory. If, moreover, one re- gards the theory of mono-anabelian log-Frobenius compatibility as corresponding to “Frobenius liftings modulo p 2 ”, then the isomorphism between Galois groups on both sides of the log-wall may be thought of as corresponding to the Frobenius TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 21 action on differentials induced by dividing the derivative of such a Frobenius lift- ing modulo p 2 by p. This correspondence between Galois groups and differentials is reminiscent of the discussion in [Mzk6], §1.3, §1.4, of the arithmetic Kodaira- Spencer morphism that arises from the [scheme-theoretic] Hodge-Arakelov theory of elliptic curves. Finally, from this point of view, it is perhaps natural to regard the mono-anabelian reconstruction algorithms of §1 as corresponding to the procedure of integrating Frobenius-invariant differentials so as to obtain canonical coordinates [i.e., “q-parameters” cf. [Mzk1], Chapter III, §1]. mono-anabelian theory p-adic theory isomorphism between Galois groups on both sides of log-wall mono-anabelian reconstruction algorithms Frobenius action on differentials arising from p 1 · derivative of mod p 2 Frobenius lifting construction of can. coords. via integration of Frobenius-invariant differentials The above discussion prompts the following question: Can one further extend the theory given in the present paper to a theory that is analogous to the theory of canonical p-adic liftings given in [Mzk1], Chapter III? It is the intention of the author to pursue the goal of developing such an “extended theory” in a future paper. Before proceeding, we note that the analogy of such a theory with the theory of canonical p-adic liftings of [Mzk1], Chapter III, may be thought of as a sort of p-adic analogue of the “geodesic flow” portion of the “rotations and geodesic flows diagram” of §I3: mono-anabelian theory p-adic theory mono-anabelian juggling of present paper, i.e., “rotations” future extended theory (?), i.e., “geodesic flows” positive characteristic [plus mod p 2 ] portion of p-adic Teichmüller theory canonical p-adic liftings in p-adic Teichmüller theory that is to say, p-adic deformations correspond to “geodesic flows”, while the positive characteristic theory corresponds to “rotations” [i.e., the theory of “mono- anabelian juggling of ,  via log” given in the present paper]. This point of view is reminiscent of the analogy between the archimedean and nonarchimedean theories discussed in Table 1 of the Introduction to [Mzk14]. In this context, it is interesting to note that this analogy between the mono- anabelian theory of the present paper and p-adic Teichmüller theory is reminiscent of various phenomena that appear in earlier papers by the author: 22 SHINICHI MOCHIZUKI (A1) In [Mzk10], Theorem 3.6, an absolute p-adic anabelian result is obtained for canonical curves as in [Mzk1] by applying the p-adic Teichmüller theory of [Mzk1]. Thus, in a certain sense [i.e., “Teichmüller =⇒ anabelian” as opposed to “anabelian =⇒ Teichmüller”], this result goes in the opposite direction to the direction of the theory of the present paper. On the other hand, this result of [Mzk10] depends on the analysis in [Mzk9], §2, of the logarithmic special fiber of a p-adic hyperbolic curve via absolute anabelian geometry over finite fields. (A2) The reconstruction of the “additive structure” via the mono-anabelian algorithms of §1 [cf. the lemma of Uchida reviewed in Proposition 1.3], which eventually leads [as discussed above], via the theory of §3, to an abstract analogue of “Frobenius liftings” [i.e., in the form of uniformizing MF -objects] is reminiscent [cf. Remark 5.10.4] of the reconstruction of the “additive structure” in [Mzk21], Corollary 2.9, via an argument analogous to an argument that may be used to show the non-existence of Frobenius liftings on p-adic hyperbolic curves [cf. [Mzk21], Remark 2.9.1]. One way to think about (A1), (A2) is by considering the following chart: p-adic Teichmüller Theory (applied to anabelian geometry) Uchida’s Lemma applied to: characteristic p special fiber Deformation Theory: canonical p-adic Frobenius liftings Future “Teichmüller-like” Extension (?) of Mono-anabelian Theory number fields, mixed char. local fields equipped with an elliptically admissible hyperbolic orbicurve analogue of Frobenius liftings in future extension (?) of mono-anabelian theory Here, the correspondence in the first non-italicized line between hyperbolic curves in positive characteristic equipped with a nilpotent ordinary indigenous bundle and number fields [and their localizations] equipped with an elliptically admissible hy- perbolic orbicurve [i.e., a hyperbolic orbicurve closely related to a once-punctured elliptic curve] has already been discussed above; the content of the “p-adic Te- ichmüller theory column” of this chart may be thought of as a summary of the content of (A1); the correspondence between this column and the “extended mono- anabelian theory” column may be regarded as a summary of the preceding dis- cussion. On the other hand, the content of (A2) may be thought of as a sort of “remarkable bridge” canonical p-adic Frobenius liftings theory of geometric uniformly toral neighborhoods =========== number fields, mixed char. local fields equipped with an elliptically admissible hyperbolic orbicurve TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 23 between the upper right-hand and lower left-hand non-italicized entries of the above chart. That is to say, the theory of (A2) [i.e., of geometric uniformly toral neigh- borhoods cf. [Mzk21], §2] is related to the upper right-hand non-italicized entry of the chart in that, like the application of “Uchida’s Lemma” represented by this entry, it provides a means for recovering the ring structure of the base field, given the decomposition groups of the closed points of the hyperbolic orbicurve. On the other hand, the theory of (A2) [i.e., of [Mzk21], §2] is related to the lower left-hand entry of the chart in that the main result of [Mzk21], §2, is obtained by an argument reminiscent [cf. [Mzk21], Remark 2.6.2; [Mzk21], Remark 2.9.1] of the argument to the effect that stable curves over rings of Witt vectors of a perfect field never admit Frobenius liftings. Note, moreover, that from the point of view of the discussion above of “arith- metic holomorphic structures”, this bridge may be thought of as a link between the elementary algebraic approach to reconstructing the “two combinatorial dimen- sions” of a ring in the fashion of Uchida’s Lemma and the “p-adic differential- geometric approach” to reconstructing p-adic ring structures in the fashion of the theory of [Mzk21], §2. Here, we observe that this “p-adic differential-geometric approach” makes essential use of the hyperbolicity of the curve under consideration. Indeed, roughly speaking, from the “Teichmüller-theoretic” point of view of the present discussion, the argument of the proof of [Mzk21], Lemma 2.6, (ii), may be summarized as follows: The nonexistence of the desired “geometric uniformly toral neighborhoods” may be thought of as a sort of nonexistence of obstructions to Teichmüller deformations of the “arithmetic holomorphic structure” that extend in an unbounded, linear fashion, like a geodesic flow or Frobenius lifting. On the other hand, the hyperbolicity of the curve under consideration implies the existence of topological obstructions i.e., in the form of “loopifica- tion” or “crushed components” [cf. [Mzk21], Lemma 2.6, (ii)] to such “unbounded” deformations of the holomorphic structure. Moreover, such “compact bounds” on the deformability of the holomorphic structure are sufficient to “trap” the holomorphic structure at a “canonical point”, which corresponds to the original holomorphic [i.e., ring] structure of interest. Put another way, this “p-adic differential-geometric interpretation of hyperbolicity” is reminiscent of the dynamics of a rubber band, whose elasticity implies that even if one tries to stretch the rubber band in an unbounded fashion, the rubber band ultimately returns to a “canonical position”. Moreover, this relationship be- tween hyperbolicity and “elasticity” is reminiscent of the use of the term “elastic” in describing certain group-theoretic aspects of hyperbolicity in the theory of [Mzk20], §1, §2. In passing, we observe that another important aspect of the theory of [Mzk21], §2, in the present context is the use of the inequality of degrees obtained by “differentiating a Frobenius lifting” [cf. [Mzk21], Remark 2.6.2]. The key importance of such degree inequalities in the theory of [Mzk21], §2, suggests, relative to the above chart, that the analogue of such degree inequalities in the theory of “the analogue of Frobenius liftings in a future extension of the mono-anabelian 24 SHINICHI MOCHIZUKI theory” could give rise to results of substantial interest in the arithmetic of number fields. The author hopes to address this topic in more detail in a future paper. Finally, we close the present Introduction to the present paper with some historical remarks. We begin by considering the following historical facts: (H1) O. Teichmüller, in his relatively short career as a mathematician, made contributions both to “complex Teichmüller theory” and to the theory of Teichmüller representatives of Witt rings two subjects that, at first glance, appear entirely unrelated to one another. (H2) In the Introduction to [Ih], Y. Ihara considers the issue of obtaining canonical p-adic liftings of certain positive characteristic hyperbolic curves equipped with a correspondence in a fashion analogous to the Serre-Tate theory of canonical liftings of abelian varieties. These two facts may be regarded as interesting precursors of the p-adic Teichmüller theory of [Mzk1], [Mzk4]. Indeed, the p-adic Teichmüller theory of [Mzk1], [Mzk4] may be regarded, on the one hand, as an analogue for hyperbolic curves of the Serre-Tate theory of canonical liftings of abelian varieties and, on the other hand, as a p-adic analogue of complex Teichmüller theory; moreover, the canonical liftings obtained in [Mzk1], [Mzk4] are, literally, “hyperbolic curve versions of Teichmüller representatives in Witt rings”. In fact, one may even go one step further to speculate that perhaps the existence of analogous complex and p-adic versions of “Teichmüller theory” should be regarded as hinting of a deeper abstract, combinatorial version of “Teichmüller theory” in a fashion that is perhaps reminiscent of the relationship of the notion of a motive to various complex or p-adic cohomology theories. It is the hope of the author that a possible “future extended theory” as discussed above i.e., a sort of “Teichmüller theory” for number fields equipped with a once-punctured elliptic curve might prove to be just such a “Teichmüller theory”. Acknowledgements: I would like to thank Yuichiro Hoshi, Fumiharu Kato, Akio Tamagawa, and Go Yamashita for countless stimulating discussions concerning the material presented in this paper and Minhyong Kim and Ben Moonen for stimulating questions related to certain of the ideas presented in this paper. Also, I feel most indebted to Yuichiro Hoshi, Go Yamashita, and Mohamed Saı̈di for their meticulous reading of and numerous comments concerning the present paper. Finally, I would like to thank Ivan Fesenko for his comments concerning the present paper. TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 25 Section 0: Notations and Conventions We shall continue to use the “Notations and Conventions” of [Mzk20], §0; [Mzk21], §0. In addition, we shall use the following notation and conventions: Numbers: In addition to the “field types” NF, MLF, FF introduced in [Mzk20], §0, we shall also consider the following field types: A(n) complex archimedean field (re- spectively, real archimedean field; archimedean field), or CAF (respectively, RAF; AF), is defined to be a topological field that is isomorphic to the field of complex numbers (respectively, the field of real numbers; either the field of real numbers or the field of complex numbers). One verifies immediately that any continuous homomorphism between CAF’s (respectively, RAF’s) is, in fact, an isomorphism of topological fields. Combinatorics: Let E be a partially ordered set. Then [cf. [Mzk16], §0] we shall denote by Order(E) the category whose objects are elements e E, and whose morphisms e 1 e 2 [where e 1 , e 2 E] are the relations e 1 e 2 . A subset E  E will be called orderwise connected if for every c E such a < c < b for some a, b E  , it follows that c E  . A partially ordered set which is isomorphic [as a partially ordered set] to an orderwise connected subset of the set of rational integers Z, equipped with its usual ordering, will be referred to as a countably ordered set. If E is a countably ordered set, then any choice of an isomorphism of E with an orderwise connected subset E  Z allows one to define [in a fashion independent of the choice of E  ], for non-maximal (respectively, non-minimal) e E [i.e., e such that there exists an f E that is > e (respectively, < e)], an element “e + 1” (respectively, “e 1”) of E. Pairs of elements of E of the form (e, e + 1) will be referred to as adjacent. An oriented graph  Γ is a graph Γ, which we shall refer to as the underlying graph of  Γ, equipped with the additional data of a total ordering, for each edge e of Γ, on the set [of cardinality 2] of branches of e [cf., e.g., [Mzk13], the discussion at the beginning of §1, for a definition of the terms “graph”, “branch”]. In this situation, we shall refer to the vertices, edges, and branches of Γ as vertices, edges, and branches of  Γ; write V(  Γ), E(  Γ), B(  Γ), respectively, for the sets of vertices, edges, and branches of  Γ. Also, whenever Γ satisfies a property of graphs [such as “finiteness”], we shall say that  Γ satisfies this property. We shall refer to the oriented graph  Γ opp obtained from  Γ by reversing the ordering on the branches of each edge as the opposite oriented graph to  Γ. A morphism of oriented graphs is defined to be a morphism of the underlying graphs [cf., e.g., [Mzk13], §1, the discussion at the beginning of §1] that is compatible with the orderings on the 26 SHINICHI MOCHIZUKI edges. Note that any countably ordered set E may be regarded as an oriented graph i.e., whose vertices are the elements of E, whose edges are the pairs of adjacent elements of E, and whose branches are equipped with the [total] ordering induced by the ordering of E. We shall refer to an oriented graph that arises from a countably ordered set as linear. We shall refer to the vertex of a linear oriented graph  Γ determined by a minimal (respectively, maximal) element of the corresponding countably ordered set as the minimal vertex (respectively, maximal vertex) of  Γ. Let  Γ be an oriented graph. Then we shall refer to as a pre-path [of length n] [where n 0 is an integer] on  Γ a morphism γ :  Γ γ  Γ, where  Γ γ is a finite linear oriented graph with precisely n edges; we shall refer to as a path [of length n] on  Γ any isomorphism class [γ] in the category of oriented graphs over  Γ of a pre-path γ [of length n]. Write Ω(  Γ) for the set [i.e., since we are working with isomorphism classes!] of paths on  Γ. If γ :  Γ γ  Γ is a pre-path on  Γ, then we shall refer to the image of the minimal (respectively, maximal) vertex of  Γ γ as the initial (respectively, terminal) vertex of γ, [γ]. Two [pre-]paths with the same initial (respectively, terminal; initial and terminal) vertices will be referred to as co-initial (respectively, co-terminal; co- verticial). If γ 1 , γ 2 are pre-paths on  Γ such the initial vertex of γ 2 is equal to the terminal vertex of γ 1 , then one may form the composite pre-path γ 2 γ 1 [in the def evident sense], as well as the composite path 2 ] 1 ] = 2 γ 1 ]. Thus, the length of γ 2 γ 1 is equal to the sum of the lengths of γ 1 , γ 2 . Next, let E Ω(  Γ) × Ω(  Γ) be a set of ordered pairs of paths on an oriented graph  Γ. Then we shall say that E is saturated if the following conditions are satisfied: (a) (Partial Inclusion of the Diagonal) If ([γ 1 ], 2 ]) E, then E contains ([γ 1 ], 1 ]) and ([γ 2 ], 2 ]). (b) (Co-verticiality) If ([γ 1 ], 2 ]) E, then 1 ], 2 ] are co-verticial. (c) (Transitivity) If ([γ 1 ], 2 ]) E and ([γ 2 ], 3 ]) E, then ([γ 1 ], 3 ]) E. (d) (Pre-composition) If ([γ 1 ], 2 ]) E and 3 ] Ω(  Γ), then ([γ 1 ]◦[γ 3 ], 2 ]◦ 3 ]) E, whenever these composite paths are defined. (e) (Post-composition) If ([γ 1 ], 2 ]) E and 3 ] Ω(  Γ), then ([γ 3 ] 1 ], 3 ] 2 ]) E, whenever these composite paths are defined. We shall say that E is symmetrically saturated if E is saturated and, moreover, satisfies the following condition: (f) (Reflexivity) If ([γ 1 ], 2 ]) E, then ([γ 2 ], 1 ]) E. Thus, the set of all co-verticial pairs of paths Covert(  Γ) Ω(  Γ) × Ω(  Γ) TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 27 is symmetrically saturated. Moreover, the property of being saturated (respectively, symmetrically saturated) is closed with respect to forming arbitrary intersections of subsets of Ω(  Γ) × Ω(  Γ). In particular, given any subset E Covert(  Γ), it makes sense to speak of the saturation (respectively, symmetric saturation) of E i.e., the smallest saturated (respectively, symmetrically saturated) subset of Covert(  Γ) containing E. Let  Γ be an oriented graph. Then we shall refer to a vertex v of  Γ as a nexus of  Γ if the following conditions are satisfied: (a) the oriented graph  Γ v obtained by removing from  Γ the vertex v, together with all of the edges that abut to v, decomposes as a disjoint union of two nonempty oriented graphs  Γ <v ,  Γ >v ; (b) every edge of  Γ that is not contained in  Γ v either runs from a vertex of  Γ <v to v or from v to a vertex of  Γ >v . In this situation, we shall refer to the oriented subgraph  Γ ≤v (respectively,  Γ ≥v ) consisting of v,  Γ <v (respectively,  Γ >v ), and all of the edges of  Γ that run to (respectively, emanate from) v as the pre-nexus portion (respectively, post-nexus portion) of  Γ. Categories: Let C, C  be categories. Then we shall use the notation Ob(C); Arr(C) to denote, respectively, the objects and arrows of C. We shall refer to a functor φ : C C  as rigid if every automorphism of φ is equal to the identity [cf. [Mzk16], §0]. If the identity functor of C is rigid, then we shall say that C is id-rigid. Let C be a category and  Γ an oriented graph. Then we shall refer to as a  Γ-diagram {A v , φ e } in C a collection of data as follows: (a) for each v V(  Γ), an object A v of C; (b) for each e E(  Γ) that runs from v 1 V(  Γ) to v 2 V(  Γ), a morphism φ e : A v 1 A v 2 of C. A morphism {A v , φ e } {A  v , φ  e } of  Γ-diagrams in C is defined to be a collection of morphisms ψ v : A v A  v for each vertex v of  Γ that are compatible with the φ e , φ  e . We shall refer to an  Γ-diagram in C as commutative if the composite morphisms determined by any co-verticial pair of paths on  Γ coincide. Write C[  Γ] for the category of commutative  Γ-diagrams in C and morphisms of  Γ-diagrams in C. If C 1 , C 2 , and D are categories, and Φ 1 : C 1 D; Φ 2 : C 2 D 28 SHINICHI MOCHIZUKI are functors, then we define the “categorical fiber product” [cf. [Mzk16], §0] C 1 × D C 2 of C 1 , C 2 over D to be the category whose objects are triples (A 1 , A 2 , α : Φ 1 (A 1 ) Φ 2 (A 2 )) where A i Ob(C i ) (for i = 1, 2), α is an isomorphism of D; and whose morphisms (A 1 , A 2 , α : Φ 1 (A 1 ) Φ 2 (A 2 )) (B 1 , B 2 , β : Φ 1 (B 1 ) Φ 2 (B 2 )) are pairs of morphisms γ i : A i B i [in C i , for i = 1, 2] such that β Φ 1 1 ) = Φ 2 2 )◦α. One verifies easily that if Φ 2 is an equivalence, then the natural projection functor C 1 × D C 2 C 1 is also an equivalence. We shall use the prefix “ind-” (respectively, “pro-”) to mean, strictly speaking, a(n) inductive (respectively, projective) system indexed by an ordered set isomor- phic to the positive (respectively, negative) integers, with their usual ordering. To simplify notation, however, we shall often denote “ind-objects” via the correspond- ing “limit objects”, when there is no fear of confusion. Let C be a category. Then we shall refer to a pair (S, A), where S Ob(C), and A Aut C (S) is a subgroup, as a pre-orbi-object of C. [Thus, we think of the pair (S, A) as representing the “stack-theoretic quotient of S by A”.] A morphism of pre-orbi-objects (S 1 , A 1 ) (S 2 , A 2 ) is an A 2 -orbit of morphisms S 1 S 2 [relative to the action of A 2 on the codomain] that is closed under the action of A 1 [on the domain]. We shall refer to as an orbi-object {(S ι , A ι ); α ι,ι  } ι,ι  ∈I any collection of data consisting of pre-orbi-objects (S ι , A ι ), which we shall refer to as representatives [of the given orbi-object], together with “gluing isomorphisms” α ι,ι  : (S ι , A ι ) (S ι  , A ι  ) of pre-orbi-objects satisfying the cocycle conditions α ι,ι  = α ι   α ι,ι  , for ι, ι  , ι  I. A morphism of orbi-objects is defined to be a collection of morphisms of pre-orbi-objects from each representative of the domain to each representative of the codomain which are compatible with the gluing isomorphisms. The category of orbi-objects associated to C is the category which we shall denote Orb(C) whose objects are the orbi-objects of C, and whose morphisms are the morphisms of orbi-objects. Thus, an object may be regarded as a pre-orbi-object whose group of automorphisms is trivial; a pre-orbi-object may be regarded as an orbi-object with precisely one representative. In particular, we obtain a natural functor C Orb(C) which is “functorial” [in the evident sense] with respect to C. TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 29 Section 1: Galois-theoretic Reconstruction Algorithms In the present §1, we apply the technique of Belyi cuspidalization developed in [Mzk21], §3, to give a group-theoretic reconstruction algorithm [cf. Theorem 1.9, Corollary 1.10] for hyperbolic orbicurves of strictly Belyi type [cf. [Mzk21], Definition 3.5] over sub-p-adic fields that is compatible with base-change of the base field. In the case of function fields, this reconstruction algorithm reduces to a much more elementary algorithm [cf. Theorem 1.11], which is valid over somewhat more general base fields, namely base fields which are “Kummer-faithful” [cf. Definition 1.5]. Let X be a hyperbolic curve over a field k. Write K X for the function field of X. Then the content of following result is a consequence of the well-known theory of divisors on algebraic curves. Proposition 1.1. (Review of Linear Systems) Suppose that X is proper, and that k is algebraically closed. Write Div(f ) for the divisor [of zeroes minus poles on X] of f K X . If E is a divisor on X, then let us write Γ × (E) = {f K X | Div(f ) + E 0} def [where we use the notation “(−) 0” to denote the effectivity of the divisor “(−)”], def l(E) = dim k (Γ(X, O X (E))). Let D be a divisor on X. Then: (i) Γ × (D) admits a natural free action by k × whenever it is nonempty; there is a natural bijection Γ × (D) Γ(X, O X (D))\{0} that is compatible with the k × - actions on either side, whenever the sets of the bijection are nonempty. (ii) The integer l(D) 0 is equal to the smallest nonnegative integer d such that there exists an effective divisor E of degree d on X for which Γ × (D E) = ∅. In particular, l(D) = 0 if and only if Γ × (D) = ∅. Proposition 1.2. (Additive Structure via Linear Systems) Let X, k be as in Proposition 1.1. Then: (i) There exist distinct points x, y 1 , y 2 X(k), together with a divisor D on X such that x, y 1 , y 2 ∈ Supp(D) [where we write Supp(D) for the support of D], such that l(D) = 2, l(D E) = 0, for any effective divisor E = e 1 + e 2 , where e 1  = e 2 , {e 1 , e 2 } {x, y 1 , y 2 }. (ii) Let x, y 1 , y 2 , D be as in (i). Then for i = 1, 2, λ k × , there ex- ists a unique element f λ,i Γ × (D) K X such that f λ,i (x) = λ, f λ,i (y i )  = 0, f λ,i (y 3−i ) = 0. (iii) Let x, y 1 , y 2 , D be as in (i); λ, μ k × such that λ/μ  = −1; f λ,1 Γ × (D) K X , f μ,2 Γ × (D) K X as in (ii). Then f λ,1 + f μ,2 Γ × (D) K X 30 SHINICHI MOCHIZUKI may be characterized as the unique element g Γ × (D) K X such that g(y 1 ) = f λ,1 (y 1 ), g(y 2 ) = f μ,2 (y 2 ). In particular, in this situation, the element λ + μ k × may be characterized as the element g(x) k × . Proof. First, we consider assertion (i). Let D be any divisor on X such that l(D) 2. By subtracting an appropriate effective divisor from D, we may assume that l(D) = 2. Then take x X(k)\Supp(D) to be any point such that O X (D) admits a global section that does not vanish at x [so l(D x) = 1]; take y 1 X(k)\(Supp(D) {x}) to be any point such that O X (D−x) admits a global section that does not vanish at y 1 [so l(D x y 1 ) = 0, which implies that l(D y 1 ) = 1]; take y 2 X(k)\(Supp(D) {x, y 1 }) to be any point such that O X (D x), O X (D y 1 ) admit global sections that do not vanish at y 2 [so l(D x y 2 ) = l(D y 1 y 2 ) = 0]. This completes the proof of assertion (i). Now assertions (ii), (iii) follow immediately from assertion (i). The following reconstruction of the additive structure from divisors and rational functions is implicit in the argument of [Uchi], §3, Lemmas 8-11 [cf. also [Tama], Lemma 4.7]. Proposition 1.3. (Additive Structure via Valuation and Evaluation Maps) Let X, k be as in Proposition 1.1. Then there exists a functorial algo- × {0} [i.e., arising from rithm for constructing the additive structure on K X the field structure of K X ] from the following data: × ; (a) the [abstract!] group K X (b) the set of [surjective] homomorphisms ×  Z} x∈X(k) V X = {ord x : K X def [so we have a natural bijection V X X(k)] that arise as valuation maps associated to points x X(k); × (c) for each homomorphism v = ord x V X , the subgroup U v K X given × by the f K X such that f (x) = 1. Here, the term “functorial” is with respect to isomorphisms [in the evident sense] of such triples [i.e., consisting of a group, a set of homomorphisms from the group to Z, and a collection of subgroups of the group parametrized by elements of this set of homomorphisms] arising from proper hyperbolic curves [i.e., “X”] over alge- braically closed fields [i.e., “k”]. × may be constructed as the inter- Proof. Indeed, first we observe that k × K X section v∈V X Ker(v). Since, for v V X , we have a direct product decomposition Ker(v) = U v × k × , the projection to k × allows us to “evaluate” elements of Ker(v) [i.e., “functions that are invertible at the point associated to v”], so as to obtain “values” of such elements k × . Next, let us observe that the set of homomorphisms TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 31 V X of (b) allows one to speak of divisors and effective divisors associated to [the × × abstract group] K X of (a). If D is a divisor associated to K X , then we may define × Γ (D) as in Proposition 1.1, and hence compute the integer l(D) as in Proposition 1.1, (ii). In particular, it makes sense to speak of data as in Proposition 1.2, (i), associated to the abstract data (a), (b), (c). Thus, by evaluating elements of vari- ous Ker(v), for v V X , we may apply the characterizations of Proposition 1.2, (ii), (iii), to construct the additive structure of k × , hence also the additive structure of × K X [i.e., by “evaluating” at various v V X ]. Remark 1.3.1. Note that if G is an abstract group, then the datum of a surjection def v : G  Z may be thought of as the datum of a subgroup H = Ker(v), together with the datum of a choice of generator of the quotient group G/H Z. Proposition 1.4. (Synchronization of Geometric Cyclotomes) Suppose that X is proper, and that k is of characteristic zero. If U X is a nonempty open subscheme, then we have a natural exact sequence of profinite groups 1 Δ U Π U G k 1 def def where we write Π U = π 1 (U ) G k = π 1 (Spec(k)) for the natural surjection of étale fundamental groups [relative to some choice of basepoints], Δ U for the kernel of this surjection. Then: def (i) Let U X be a nonempty open subscheme, x X(k)\U (k), U x = X\{x} X. Then the inertia group I x of x in Δ U is naturally isomorphic  to Z(1); the kernels of the natural surjections Δ U  Δ U x , Π U  Π U x are topolog- ically normally generated by the inertia groups of points of U x \U [each of which is  naturally isomorphic to Z(1)]. (ii) Let x, U x be as in (i). Then we have a natural exact sequence of profinite groups 1 I x Δ c-cn U x Δ X 1 where we write Δ U x  Δ c-cn U x for the maximal cuspidally central quotient of Δ U x [i.e., the maximal intermediate quotient Δ U x  Q  Δ X such that Ker(Q  Δ X ) lies in the center of Q cf. [Mzk19], Definition 1.1, (i)]. Moreover, applying the differential of the “E 2 -term” of the Leray spectral sequence associated to this group extension to the element  = Hom(I x , I x ) = H 0 X , H 1 (I x , I x )) 1 Z yields an element H 2 X , H 0 (I x , I x )) = Hom(M X , I x ), where we write def  Z)  M X = Hom(H 2 X , Z), [cf. the discussion at the beginning of [Mzk19], §1]; this last element corresponds to the natural isomorphism M X I x 32 SHINICHI MOCHIZUKI  [relative to the well-known natural identifications of I x , M X with Z(1) cf., e.g., (i) above; [Mzk19], Proposition 1.2, (i)]. In particular, this yields a “purely group-theoretic algorithm” [cf. Remark 1.9.8 below for more on the meaning of this terminology] for constructing this isomorphism from the surjection Δ U x  Δ X . Proof. Assertion (i) is well-known [and easily verified from the definitions]. Asser- tion (ii) follows immediately from [Mzk19], Proposition 1.6, (iii). Definition 1.5. Let k be a field of characteristic zero, k an algebraic closure def of k, G k = Gal(k/k). Then we shall say that k is Kummer-faithful (respectively, torally Kummer-faithful) if, for every finite extension k H k of k, where we write def H = Gal(k/k H ) G k , and every semi-abelian variety (respectively, every torus) A over k H , either of the following two equivalent conditions is satisfied: (a) We have N · A(k H ) = {0} N ≥1 where N ranges over the positive integers. (b) The associated Kummer map A(k H ) H 1 (H, Hom(Q/Z, A(k))) is an injection. [To verify the equivalence of (a) and (b), it suffices to consider, on the étale site of Spec(k H ), the long exact sequences in étale cohomology associated to the exact N · sequences 0 −→ N A −→ A −→ A −→ 0 arising from multiplication by positive integers N .] Remark 1.5.1. In the notation of Definition 1.5, suppose that k is a torally Kummer-faithful field, l a prime number. Then it follows immediately from the injectivity of the Kummer map associated to G m over any finite extension of k that contains a primitive l-th root of unity that the cyclotomic character χ l : G k Z × l has open image [cf. the notion of “l-cyclotomic fullness” discussed in [Mzk20], Lemma 4.5]. In particular, it makes sense to speak of the “power-equivalence class of χ l [cf. [Mzk20], Lemma 4.5, (ii)] among characters G k Z × l i.e., the equivalence class with respect to the equivalence relation ρ 1 ρ 2 [for characters N N ρ 1 , ρ 2 : G k Z × l ] defined by the condition that ρ 1 = ρ 2 for some positive integer N . Remark 1.5.2. By considering the Weil restrictions of semi-abelian varieties or tori over finite extensions of k to k, one verifies immediately that one obtains an equivalent definition of the terms “Kummer-faithful” and “torally Kummer- faithful” if, in Definition 1.5, one restricts k H to be equal to k. Remark 1.5.3. In the following discussion, if k is a field, then we denote the subgroup of roots of unity of k × by μ(k) k × . TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 33 (i) Let k be a(n) [not necessarily finite!] algebraic field extension of a number field such that there exists a nonarchimedean prime of k that is unramified over some number field contained in k, and, moreover, for every finite extension k of k, μ(k ) is finite. Then I claim that: k is torally Kummer-faithful. Indeed, since [as one verifies immediately] any finite extension of k satisfies the same hypotheses as k, one verifies immediately that it suffices to show that N (k × ) N = {1} [where N ranges over the positive integers]. Let f N (k × ) N . If f μ(k), then the assumption concerning μ(k) implies immediately that f = 1; thus, we may assume without loss of generality that f ∈ μ(k). But then there exists a nonarchimedean prime p of k that is unramified over some number field contained in k. In particular, if we write k p for the completion of k at p and p for the residue characteristic of p, then k p embeds into a finite extension of the quotient field of the ring of Witt vectors of an algebraic closure of F p . Thus, the fact that f admits arbitrary p-power roots in k p yields a contradiction. This completes the proof of the claim. (ii) It follows immediately from the definitions that “Kummer-faithful =⇒ torally Kummer-faithful”. On the other hand, as was pointed out to the author by A. Tamagawa, one may construct an example of a field which is torally Kummer- faithful, but not Kummer-faithful, as follows: Let E be an elliptic curve over a number field k 0 that admits complex multiplication by Q( −1) and, moreover, has good reduction at every nonarchimedean prime of k 0 . Let k be an algebraic closure def of k 0 , G k 0 = Gal(k/k 0 ), p a prime number 1 (mod 4). Write V for the p-adic Tate module associated to E. Thus, the G k 0 -module V decomposes [since p 1 (mod 4)] into a direct sum W W  of submodules W, W  V of rank one. Write χ : G k 0 Z × p for the character determined by W . Thus, [as is well-known] χ is unramified over every nonarchimedean prime of k 0 of residue characteristic l  = p, as well as over some nonarchimedean prime of k 0 of residue characteristic p. But one verifies immediately [for instance, by considering ramification over Q] that this implies that, if we write k for the extension field of k 0 determined by the kernel of χ, then μ(k) is finite. Thus, since any finite extension of k 0 satisfies the same hypotheses as k 0 , we conclude that k satisfies the hypotheses of (i), so k is torally Kummer-faithful. On the other hand, [by the definition of χ, W , V !] the Kummer map on E(k) is not injective, so k is not Kummer-faithful. Remark 1.5.4. (i) Observe that every sub-p-adic field k [cf. [Mzk5], Definition 15.4, (i)] is Kummer-faithful, i.e., “sub-p-adic =⇒ Kummer-faithful”. Indeed, to verify this, one reduces immediately, by base-change, to the case where k is a finitely generated extension of an MLF, which may be thought of as the function field of a variety over an MLF. Then by restricting to various closed points of this variety, one reduces to the case where k itself is an MLF. On the other hand, if k, hence also k H [cf. the notation of Definition 1.5], is a finite extension of Q p , then A(k H ) is an extension of a finitely generated Z-module by a compact abelian p-adic Lie group, hence contains an open subgroup that is isomorphic to a finite product of copies of Z p . In particular, the condition of Definition 1.5, (a), is satisfied. 34 SHINICHI MOCHIZUKI (ii) A similar argument to the argument of (i) shows that every finitely gen- erated extension of a Kummer-faithful field (respectively, torally Kummer-faithful field) is itself Kummer-faithful (respectively, torally Kummer-faithful field). (iii) On the other hand, observe that if, for instance, I is an infinite set, then the def field k = Q p (x i ) i∈I [which is not a finitely generated extension of Q p ] constitutes an example of a Kummer-faithful field which is not sub-p-adic. Indeed, if, for H, A as in Definition 1.5, 0  = f A(k H ) lies in the kernel of the associated Kummer map, then observe that there exists some finite subset I  I such that if we set def  k  = Q p (x i ) i∈I  , then, for some finite extension k H k H of k  , we may assume   , that f A  (k H ) A(k H ), that A descends to a semi-abelian variety A  over k H def   (x i ) i∈I  , where we set I  = I\I  . Since k H is algebraically closed and that k H = k H in k H , it thus follows that all roots of f defined over k H are in fact defined over   . Thus, the existence of f contradicts the fact that the sub-p-adic field k H is k H Kummer-faithful. Finally, to see that k is not sub-p-adic, suppose that k K, where K is a finitely generated extension of an MLF K 0 of residue characteristic p 0 such that K 0 is algebraically closed in K. Let l  = p, p 0 be a prime number. Then Q p k = (k × ) l K = N def (K × ) l K 0 N def l N l N where one verifies immediately that the additive group generated by k (respec- tively, K ) in k (respectively, K) forms a compact open neighborhood of 0 in Q p (respectively, K 0 ). In particular, it follows that the inclusion k → K determines a continuous homomorphism of topological fields Q p → K 0 . But this implies imme- diately that p 0 = p, and that Q p → K 0 is a Q p -algebra homomorphism. Thus, the theory of transcendence degree yields a contradiction [for instance, by considering the morphism on Kähler differentials induced by k → K]. (iv) One verifies immediately that the generalized sub-p-adic fields of [Mzk8], Definition 4.11, are not, in general, torally Kummer-faithful. Proposition 1.6. (Kummer Classes of Rational Functions) In the situ- ation of Proposition 1.4, suppose further that k is a Kummer-faithful field. If U X is a nonempty open subscheme, then let us write × ) H 1 U , M X ) κ U : Γ(U, O U  where M X is as in Proposition 1.4, (ii) for the associated Kummer = Z(1) map [cf., e.g., the discussion at the beginning of [Mzk19], §2]. Also, for d Z, let us write J d Spec(k) for the connected component of the Picard scheme of X Spec(k) that parametrizes line bundles of degree d [cf., e.g., the discussion def def preceding [Mzk19], Proposition 2.2]; J = J 0 ; Π J d = π 1 (J d ). [Thus, we have a natural morphism X J 1 that sends a point of X to the line bundle of degree 1 associated to the point; this morphism induces a surjection Π X  Π J 1 on étale fundamental groups whose kernel is equal to the commutator subgroup of Δ X .] Then: (i) The Kummer map κ U is injective. TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 35 (ii) For x X(k), write s x : G k Π X for the associated section [well- defined up to conjugation by Δ X ], t x : G k Π J 1 for the composite of s x with the natural surjection Π X  Π J 1 . Then for any divisor D of degree d on X such that Supp(D) X(k), forming the appropriate Z-linear combination of “t x ’s” for x Supp(D) [cf., e.g., the discussion preceding [Mzk19], Proposition 2.2] yields a section t D : G k Π J d ; if, moreover, d = 0, then t D : G k Π J coincides [up to conjugation by Δ X ] with the section determined by the identity element J(k) if and only if the divisor D is principal. (iii) Suppose that U = X\S, where S X(k) is a finite subset. Then restrict- ing cohomology classes of Π U to the various I x [cf. Proposition 1.4, (i)], for x S, yields a natural exact sequence   × 1  1 (k ) H U , M X ) Z x∈S  via the isomorphism I x (I , M X ) with Z M X of where we identify Hom  Z x Proposition 1.4, (ii); (k × ) denotes the profinite completion of k × . Moreover, the × image [via κ U ] of Γ(U, O U ) in H 1 U , M X )/(k × ) is equal to the inverse image in H 1 U , M X )/(k × ) of the submodule of      Z Z x∈S x∈S determined by the principal divisors [with support in S]. Proof. Assertion (i) follows immediately [by restricting to smaller and smaller “U ’s”] from the fact [cf. Remark 1.5.4, (ii)] that since k is [torally] Kummer- faithful, so is the function field K X of X. Assertion (ii) follows from the argument of [Mzk19], Proposition 2.2, (i), together with the assumption that k is Kummer- faithful. As for assertion (iii), just as in the proof of [Mzk19], Proposition 2.1, (ii), to verify assertion (iii), it suffices to verify that H 0 (G k , Δ ab X ) = 0; but, in light of and the torsion points of the Jacobian J, the well-known relationship between Δ ab X 0 ab the fact that H (G k , Δ X ) = 0 follows immediately from our assumption that k is Kummer-faithful [cf. the argument applied to G m in Remark 1.5.1]. Definition 1.7. Suppose that k is of characteristic zero. Let k be an algebraic closure of k; write k NF k for the [“number field”] algebraic closure of Q in k. def (i) We shall say that X is an NF-curve if X k = X × k k is defined over k NF [cf. Remark 1.7.1 below]. (ii) Suppose that X is an NF-curve. Then we shall refer to points of X(k) (respectively, rational functions on X k ; constant rational functions on X k [i.e., which arise from elements of k]) that descend to k NF [cf. Remark 1.7.1 below] as NF-points of (respectively, NF-rational functions on; NF-constants on) X k . Remark 1.7.1. Suppose that X is of type (g, r). Then observe that X is an NF-curve if and only if the k-valued point of the moduli stack of hyperbolic curves 36 SHINICHI MOCHIZUKI of type (g, r) over Q determined by X arises, in fact, from a k NF -valued point. In particular, one verifies immediately that if X is an NF-curve, then the descent data of X k from k to k NF is unique. Proposition 1.8. (Characterization of NF-Constants and NF-Rational Functions) In the situation of Proposition 1.6, (iii), suppose further that U [hence also X] is an NF-curve. Write P U H 1 U , M X ) for the inverse image of the submodule of      Z Z x∈S x∈S determined by the cuspidal principal divisors [i.e., principal divisors supported on the cusps] cf. Proposition 1.6, (iii). Then: (i) A class η P U is the Kummer class of a nonconstant NF-rational function if and only if there exist a positive multiple η of η and NF-points x i U (k x ), where i = 1, 2, and k x is a finite extension of k, such that the coho- mology classes def η | x i = s x i ) H 1 (G k x , M X ) where we write s x i : G k x Π U for the [outer] homomorphism determined by x i [cf. the notation of Proposition 1.6, (ii)] satisfy η | x 1 = 0 [i.e., = 1, if one works multiplicatively], η | x 2  = 0. × ). (ii) Suppose that there exist nonconstant NF-rational functions Γ(U, O U 1 × Then a class η P U H (G k , M X ) = (k ) [cf. the exact sequence of Proposition 1.6, (iii)] is the Kummer class of an NF-constant k × if and only if there exist a × nonconstant NF-rational function f Γ(U, O U ) and an NF-point x U (k x ), where k x is a finite extension of k, such that κ U (f )| x = η| G kx H 1 (G k x , M X ) where we use the notation “| x as in (i). Proof. Suppose that X k descends to a hyperbolic curve X NF over k NF . Then [since k NF is algebraically closed] any nonconstant rational function on X NF determines a morphism X NF P 1 k such that the induced map X NF (k NF ) P 1 k (k NF ) is NF NF surjective. In light of this fact [cf. also the fact that U is also assumed to be an NF-curve], assertions (i), (ii) follow immediately from the definitions. Now, by combining the “reconstruction algorithms” given in the various results discussed above, we obtain the main result of the present §1. Theorem 1.9. (The NF-portion of the Function Field via Belyi Cus- pidalization over Sub-p-adic Fields) Let X be a hyperbolic orbicurve of TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 37 strictly Belyi type [cf. [Mzk21], Definition 3.5] over a sub-p-adic field [cf. [Mzk5], Definition 15.4, (i)] k, for some prime p; k an algebraic closure of k; k NF k the algebraic closure of Q in k; 1 Δ X Π X G k 1 def def where Π X = π 1 (X) G k = Gal(k/k) denotes the natural surjection of étale fundamental groups [relative to some choice of basepoints], and Δ X denotes the kernel of this surjection the resulting extension of profinite groups. Then there exists a functorial “group-theoretic” algorithm [cf. Remark 1.9.8 below for more on the meaning of this terminology] for reconstructing the “NF-portion of the function field” of X from the extension of profinite groups 1 Δ X Π X G k 1; this algorithm consists of the following steps: (a) One constructs the various surjections Π U  Π Y where Y is a hyperbolic [NF-]curve that arises as a finite étale covering of X; U Y is an open subscheme obtained by removing an arbitrary def def finite collection of NF-points; Π U = π 1 (U ); Π Y = π 1 (Y ) Π X via the technique of “Belyi cuspidalization”, as described in [Mzk21], Corollary 3.7, (a), (b), (c). Here, we note that by allowing U to vary, we obtain a “group-theoretic” construction of Π U equipped with the collection of subgroups that arise as decomposition groups of NF-points. (b) One constructs the natural isomorphisms def I z μ  U ) = M Z Z where U Y X is as in (i), Y is of genus 2, Z is the canonical compactification of Y , the points of Z\U are all rational over the base field k Z of Z, z (Z\U )(k Z ) via the technique of Proposition 1.4, (ii). (c) For U Y Z as in (b), one constructs the subgroup P U H 1 U , μ  U )) Z determined by the cuspidal principal divisors via the isomorphisms of (b) and the characterization of principal divisors given in Proposition 1.6, (ii) [cf. also the decomposition groups of (a); Proposition 1.6, (iii)]. (d) For U Y Z, k Z as in (b), one constructs the subgroups × 1 k NF K Z × NF → lim U )) −→ H V , μ  Z V where V ranges over the open subschemes obtained by removing finite collections of NF-points from Z × k Z k  , for k  a finite extension of k Z ; def Π V = π 1 (V ); K Z NF is the function field of the curve Z NF obtained by 38 SHINICHI MOCHIZUKI descending Z × k Z k to k NF ; the “→” arises from the Kummer map via the subgroups of (c) and the characterizations of Kummer classes of nonconstant NF-rational functions and NF-constants given in Proposition 1.8, (i), (ii) [cf. also the decomposition groups of (a)]. (e) One constructs the additive structure on × k NF  {0}; K Z × NF  {0} [notation as in (d)] by applying the functorial algorithm of Proposition 1.3 to the data of the form described in Proposition 1.3, (a), (b), (c), arising from the construction of (d) [cf. also the decomposition groups of (a), the isomorphisms of (b)]. Finally, the asserted “functoriality” is with respect to arbitrary open injective homomorphisms of extensions of profinite groups [cf. also Remark 1.10.1 below], as well as with respect to homomorphisms of extensions of profinite groups arising from a base-change of the base field [i.e., k]. Proof. The validity of the algorithm asserted in Theorem 1.9 is immediate from the various results cited in the statement of this algorithm. Remark 1.9.1. When k is an MLF [cf. [Mzk20], §0], one verifies immediately that one may give a tempered version of Theorem 1.9 [cf. [Mzk21], Remark 3.7.1], in which the profinite étale fundamental group Π X is replaced by the tempered funda- mental group of X [and the expression “profinite group” is replaced by “topological group”]. Remark 1.9.2. When k is an MLF or NF [cf. [Mzk20], §0], the “extension of profinite groups 1 Δ X Π X G k 1” that appears in the input data for the algorithm of Theorem 1.9 may be replaced by the single profinite group Π X [cf. [Mzk20], Theorem 2.6, (v), (vi)]. A similar remark applies in the tempered case discussed in Remark 1.9.1. Remark 1.9.3. Note that unlike the case with k NF , K Z NF , the algorithm of Theorem 1.9 does not furnish a means for reconstructing k, K Z in general cf. Corollary 1.10 below concerning the case when k is an MLF. Remark 1.9.4. Suppose that k is an MLF. Then G k , which is of cohomological dimension 2 [cf. [NSW], Theorem 7.1.8, (i)], may be thought of as having one rigid dimension and one non-rigid dimension. Indeed, the maximal unramified quotient  G k  G unr = Z k is generated by the Frobenius element, which may be characterized by an entirely group-theoretic algorithm [hence is preserved by isomorphisms of absolute Galois groups of MLF’s cf. [Mzk9], Proposition 1.2.1, (iv)]; thus, this quotient G k  TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 39  may be thought of as a “rigid dimension”. On the other hand, the G unr = Z k dimension of G k represented by the inertia group in I k G k [which, as is well-known, is of cohomological dimension 1] is “far from rigid” a phenomenon that may be seen, for instance, in the existence [cf., e.g., [NSW], the Closing Remark preceding Theorem 12.2.7] of isomorphisms of absolute Galois groups of MLF’s which fail [equivalently cf. [Mzk20], Corollary 3.7] to be “RF- preserving”, “uniformly toral”, or “geometric”. By contrast, it is interesting to observe that: The “group-theoretic” algorithm of Theorem 1.9 shows that the condition of being “coupled with Δ X [i.e., via the extension determined by Π X ] has the effect of rigidifying both of the 2 dimensions of G k [cf. also Corollary 1.10 below]. This point of view will be of use in our development of the archimedean theory in §2 below [cf., e.g., Remark 2.7.3 below]. Remark 1.9.5. (i) Note that the functoriality with respect to isomorphisms of the algorithm of Theorem 1.9 may be regarded as yielding a new proof of the “profinite absolute version of the Grothendieck Conjecture over number fields” [cf., e.g., [Mzk15], The- orem 3.4] that does not logically depend on the theorem of Neukirch-Uchida [cf., e.g., [Mzk15], Theorem 3.1]. Moreover, to the author’s knowledge: The technique of Theorem 1.9 yields the first logically independent proof of a consequence of the theorem of Neukirch-Uchida that involves an explicit construction of the number fields involved. Put another way, the algorithm of Theorem 1.9 yields a proof of a consequence of the theorem of Neukirch-Uchida on number fields in the style of Uchida’s work on function fields in positive characteristic [i.e., [Uchi]] cf., especially, Proposition 1.3. (ii) One aspect of the theorem of Neukirch-Uchida is that its proof relies es- sentially on the data arising from the decomposition of primes in finite extensions of a number field i.e., in other words, on the “global address” of a prime among all the primes of a number field. Such a “global address” is manifestly annihilated by the operation of localization at the prime under consideration. In particular, the crucial functoriality of Theorem 1.9 with respect to change of base field [e.g., from a number field to a nonarchimedean completion of the number field] is an- other reflection of the way in which the nature of the proof of Theorem 1.9 over number fields differs quite fundamentally from the essentially global proof of the theorem of Neukirch-Uchida [cf. also Remark 3.7.6, (iii), (v), below]. This “crucial functoriality” may also be thought of as a sort of essential independence of the algorithms of Theorem 1.9 from both methods which are essentially global in nature 40 SHINICHI MOCHIZUKI [such as methods involving the “global address” of a prime] and methods which are essentially local in nature [such as methods involving p-adic Hodge theory cf. Remark 3.7.6, (iii), (v), below]. This point of view concerning the “essential independence of the base field” is developed further in Remark 1.9.7 below. Remark 1.9.6. By combining the theory of the present §1 with the theory of [Mzk21], §1 [cf., e.g., [Mzk21], Corollary 1.11 and its proof], one may obtain “functorial group-theoretic reconstruction algorithms”, in a number of cases, for finite étale coverings of configuration spaces associated to hyperbolic curves. We leave the routine details to the interested reader. Remark 1.9.7. One way to think of the construction algorithm in Theorem 1.9 of the “NF-portion of the function field” of a hyperbolic orbicurve of strictly Belyi type over a sub-p-adic field is the following: The algorithm of Theorem 1.9 may be thought of as a sort of complete “combinatorialization” independent of the base field! of the [algebro-geometric object constituted by the] orbicurve under considera- tion. This sort of “combinatorialization” may be thought of as being in a similar vein albeit much more technically complicated! to the “combinatorialization” of a category of finite étale coverings of a connected scheme via the notion of an abstract Galois category, or the “combinatorialization” of certain aspects of the commutative algebra of “normal rings with toral singularities” via the abstract monoids that appear in the theory of log regular schemes [cf. also the Introduction of [Mzk16] for more on this point of view]. Remark 1.9.8. Typically in discussions of anabelian geometry, the term “group- theoretic” is applied to a property or construction that is preserved by the isomor- phisms [or homomorphisms] of fundamental groups under consideration [cf., e.g., [Mzk5]]. By contrast, our use of this term is intended in a stronger sense. That is to say: We use the term “group-theoretic algorithm” to mean that the algorithm in question is phrased in language that only depends on the topological group structure of the fundamental group under consideration. [Thus, the more “classical” use [e.g., in [Mzk5]] of the term “group-theoretic” cor- responds, in our discussion of “group-theoretic algorithms”, to the functoriality e.g., with respect to isomorphisms of some type of the algorithm.] In particu- lar, one fundamental difference between the approach usually taken to anabelian geometry and the approach taken in the present paper is the following: The “classical” approach to anabelian geometry, which we shall refer to as bi-anabelian, centers around a comparison between two geomet- ric objects [e.g., hyperbolic orbicurves] via their [arithmetic] fundamental TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 41 groups. By contrast, the theory of the present paper, which we shall refer to as mono-anabelian, centers around the task of establishing “group- theoretic algorithms” i.e., “group-theoretic software” that require as input data only the [arithmetic] fundamental group of a single geometric object. Thus, it follows formally that “mono-anabelian” =⇒ “bi-anabelian”. On the other hand, if one is allowed in one’s algorithms to introduce some fixed reference model of the geometric object under consideration, then the task of establishing an “algorithm” may, in effect, be reduced to “comparison with the fixed reference model”, i.e., reduced to some sort of result in “bi-anabelian geometry”. That is to say, if one is unable to settle the issue of ruling out the use of such models, then there remains the possibility that ? “bi-anabelian” =⇒ “mono-anabelian”. We shall return to this crucial issue in §3 below [cf., especially, Remark 3.7.3]. Remark 1.9.9. As was pointed out to the author by M. Kim, one may also think of the algorithms of a result such as Theorem 1.9 as suggesting an approach to solving the problem of characterizing “group-theoretically” those profinite groups Π that occur [i.e., in Theorem 1.9] as a “Π X ”. That is to say, one may try to obtain such a characterization by starting with, say, an arbitrary slim profinite group Π and then proceeding to impose “group-theoretic” conditions on Π corresponding to the various steps of the algorithms of Theorems 1.9 i.e., conditions whose content consists of minimal assumptions on Π that are necessary in order to execute each step of the algorithm. Corollary 1.10. (Reconstruction of the Function Field for MLF’s) Let X be a hyperbolic orbicurve over an MLF k [cf. [Mzk20], §0]; k an algebraic closure of k; k NF k the algebraic closure of Q in k; 1 Δ X Π X G k 1 def def where Π X = π 1 (X) G k = Gal(k/k) denotes the natural surjection of étale fundamental groups [relative to some choice of basepoints], and Δ X denotes the kernel of this surjection the resulting extension of profinite groups. Then: (i) There exists a functorial “group-theoretic” algorithm for reconstruct-  (G k )) Z [cf. (a) below], together with ing the natural isomorphism H 2 (G k , μ  Z 1 ab  [cf. (b) below] from the (G k )) G k  Z the natural surjection H (G k , μ  Z profinite group G k , as follows: (a) Write: def ab μ Q/Z (G k ) = lim −→ (H ) tors ; H def μ  (G k ) = Hom(Q/Z, μ Q/Z (G k )) Z 42 SHINICHI MOCHIZUKI where H ranges over the open subgroups of G k ; the notation “(−) tors denotes the torsion subgroup of the abelian group in parentheses; the arrows of the direct limit are induced by the Verlagerung, or transfer, map [cf. the discussion preceding [Mzk9], Proposition 1.2.1; the proof of [Mzk9], (G k ) is Proposition 1.2.1]. [Thus, the underlying module of μ Q/Z (G k ), μ  Z unaffected by the operation of passing from G k to an open subgroup of G k .] Then one constructs the natural isomorphism  H 2 (G k , μ  (G k )) Z Z “group-theoretically” from G k via the algorithm described in the proof of [Mzk9], Proposition 1.2.1, (vii). (b) By applying the isomorphism of (a) [and the cup-product in group coho- mology], one constructs the surjection unr  (G k )) G ab Z H 1 (G k , μ  k  G Z determined by the Frobenius element in the maximal unramified quo- tient G unr of G k via the “group-theoretic” algorithm described in the proof of [Mzk9], Proposition 1.2.1, (ii), (iv). Here, the asserted “functoriality” is with respect to arbitrary injective open homomorphisms of profinite groups [cf. also Remark 1.10.1, (iii), below]. (ii) By applying the functorial “group-theoretic” algorithm of [Mzk20], Lemma 4.5, (v), to construct the decomposition groups of cusps in Π X , one ob- X ) as in Proposition 1.4, (ii); Theorem 1.9, (b) [cf. also tains a Π X -module μ  Z Remark 1.10.1, (ii), below]. Then there exists a functorial “group-theoretic” algorithm for reconstructing the natural isomorphism μ  (G k ) μ  X ) [cf. Z Z (c) below; Remark 1.10.1 below] and the image of a certain Kummer map [cf. (d) below] from the profinite group Π X [cf. Remark 1.9.2], as follows: (c) One constructs the natural isomorphism [cf., e.g., [Mzk12], Theorem 4.3] μ  (G k ) μ  X ) Z Z thought of as an element of the quotient H 1 (G k , μ  X ))  Hom(μ  (G k ), μ  X )) Z Z Z determined by the surjection of (b) as the unique topological genera- tor of Hom(μ  (G k ), μ  X )) that is contained in the “positive rational Z Z structure” [arising from various J ab , for J Δ X an open subgroup] of [Mzk9], Lemma 2.5, (i) [cf. also [Mzk9], Lemma 2.5, (ii)]. (d) One constructs the image of the Kummer map X )) → H 1 X , μ  X )) k × → H 1 (G k , μ  Z Z TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 43 as the inverse image of the subgroup generated by the Frobenius element  X )) H 1 (G k , μ  (G k )) G ab via the surjection H 1 (G k , μ  k  Z of Z Z (b) [cf. also the isomorphism of (c)]. (d  ) Alternatively, if X is of strictly Belyi type [so that we are in the situ- ation of Theorem 1.9], then one may construct the image of the Kummer map of (d) without applying the isomorphism of (c) as the com- × )) k pletion of H 1 (G k , μ  X NF [cf. Theorem 1.9, (e)] with respect to Z × X )) k NF ) {0} [relative to the the valuation on the field (H 1 (G k , μ  Z additive structure of Theorem 1.9, (e)] determined by the subring of this ×  ))  Z) k field generated by the intersection Ker(H 1 (G k , μ  X NF Z  × , where “” is the surjection of (b), considered up to multiplication by Z an object which is independent of the isomorphism of (c). Here, the asserted “functoriality” is with respect to arbitrary open injective homomorphisms of extensions of profinite groups cf. Remark 1.10.1 below. (iii) Suppose further that X is of strictly Belyi type [so that we are in the situation of Theorem 1.9]. Then there exists a functorial “group-theoretic” algorithm for reconstructing the function field K X of X from the profinite group Π X [cf. Remark 1.9.2], as follows: (e) One constructs the decomposition groups in Π X of arbitrary closed points of X by approximating such points by NF-points of X [whose decomposition groups have already been constructed, in Theorem 1.9, (a)], via the equivalence of [Mzk12], Lemma 3.1, (i), (iv). (f) For S a finite set of closed points of X, one constructs the associated “maximal abelian cuspidalization” Π c-ab U S def of U S = X\S via the algorithm of [Mzk19], Theorem 2.1, (i) [cf. also [Mzk19], Theorem 1.1, (iii), as well as Remark 1.10.4, below]. Moreover, by applying the approximation technique of (e) to the Belyi cuspidal- izations of Theorem 1.9, (a), one may construct the Green’s trivial- izations [cf. [Mzk19], Definition 2.1; [Mzk19], Remark 15] for arbitrary pairs of closed points of X such that one point of the pair is an NF-point; in particular, one may construct the liftings to Π c-ab U S [from Π X ] of de- composition groups of NF-points. (g) By applying the “maximal abelian cuspidalizations” Π c-ab U S of (f ), together with the characterization of principal divisors given in Proposition 1.6, (ii) [cf. also the decomposition groups of (e)], one constructs the subgroup X )) ( X ))) P U S H 1 c-ab = H 1 U S , μ  U S , μ  Z Z [cf. [Mzk19], Proposition 2.1, (i), (ii)] determined by the cuspidal prin- cipal divisors via the isomorphisms of Theorem 1.9, (b). Then the 44 SHINICHI MOCHIZUKI image of the Kummer map in P U S may be constructed as the collec- X )) where tion of elements of P U S whose restriction H 1 (G k  , μ  Z G k  G k is an open subgroup corresponding to a finite extension k  k of k to a decomposition group of some NF-point [cf. (f )] is contained in (k  ) × ((k  ) × ) H 1 (G k  , μ  X )) [cf. (d) or, alternatively,(d’)]. Z (h) One constructs the additive structure on [the image cf. (d) of ] k × {0} as the unique continuous extension of the additive structure × on (k × k NF ) {0} constructed in Theorem 1.9, (e). One constructs the image of the Kummer map × 1 c-ab K X → lim X )) −→ H U S , μ  Z S by letting S as in (g) vary. One constructs the additive structure on × K X {0} as the unique additive structure compatible, relative to the oper- ation of restriction to decomposition groups of NF-points [cf. (f )], with the additive structures on the various (k  ) × {0}, for k  k a finite exten- × sion of k. Also, one may construct the restrictions of elements of K X to decomposition groups not only of NF-points, but also of arbitrary closed points of X, by approximating as in (e); this allows one [by letting k vary among finite extensions of k in k] to give an alternative construction × {0} by applying Proposition 1.3 directly of the additive structure on K X [i.e., over k, as opposed to k NF ]. Here, the asserted “functoriality” is with respect to arbitrary open injective homomorphisms of profinite groups [i.e., of “Π X ”] cf. Remark 1.10.1 below. Proof. The validity of the algorithms asserted in Corollary 1.10 is immediate from the various results cited in the statement of these algorithms. Remark 1.10.1. (i) In general, the functoriality of Theorem 1.9, Corollary 1.10, when applied to the operation of passing to open subgroups of Π X , is to be understood in the sense of a “compatibility”, relative to dividing the usual functorially induced morphism X )’s” by a factor given by the index of the subgroups of Δ X that arise on “μ  Z from the open subgroups of Π X under consideration [cf., e.g., [Mzk19], Remark 1]. U )” in Theorem 1.9, (b), (ii) In fact, strictly speaking, the definition of “μ  Z is only valid if U is a hyperbolic curve of genus 2; nevertheless, one may extend this definition to the case where U is an arbitrary hyperbolic orbicurve precisely by passing to coverings and applying the “functoriality/compatibility” discussed in (i). We leave the routine details to the reader.  (G k )) Z of (iii) In a similar vein, note that the isomorphism H 2 (G k , μ  Z Corollary 1.10, (a), is functorial in the sense that it is compatible with the result of dividing the usual functorially induced morphism by a factor given by the index of the open subgroups of G k under consideration. TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 45 Remark 1.10.2. Just as was the case with Theorem 1.9, one may give a tempered version of Corollary 1.10 cf. Remark 1.9.1. Remark 1.10.3. (i) The isomorphism of Corollary 1.10, (c), may be thought of as a sort of “synchronization of [arithmetic and geometric] cyclotomes”, in the style of the “synchronization of cyclotomes” given in the final display of Proposition 1.4, (ii). (ii) One may construct the natural isomorphism 1 X )) G ab k H (G k , μ  Z by applying the displayed isomorphism of Corollary 1.10, (c), to the inverse of the first displayed isomorphism of Corollary 1.10, (b). By applying this natural isomor- phism to various open subgroups of G k , we thus obtain yet another isomorphism of cyclotomes × def κ μ  (G ) μ ) = Hom(Q/Z, κ(k k X NF )) Z  Z × × where we write κ(k NF ) for the image of k NF in 1 lim U )) −→ H V , μ  Z V via the inclusion induced by the Kummer map in the display of Theorem 1.9, (d). Remark 1.10.4. Here, we take the opportunity to correct an unfortunate misprint  X, Z Y  Y in the proof of [Mzk19], Theorem 1.1, (iii). The phrase “Z X are diagonal coverings” that appears at the beginning of this proof should read  X × X, Z Y  Y × Y are diagonal coverings”. “Z X Finally, we conclude the present §1 by observing that the techniques developed in the present §1 may be intepreted as implying a very elementary semi-absolute birational analogue of Theorem 1.9. Theorem 1.11. (Semi-absolute Reconstruction of Function Fields of Curves over Kummer-faithful Fields) Let X be a smooth, proper, geometrically connected curve of genus g X over a Kummer-faithful field k; K X the function def field of X; η X = Spec(K X ); k an algebraic closure of k; 1 Δ η X Π η X G k 1 def def where Π η X = π 1 X ) G k = Gal(k/k) denotes the natural surjection of étale fundamental groups [relative to some choice of basepoints], and Δ η X denotes the kernel of this surjection the resulting extension of profinite groups. Then Δ η X , Π η X , and G k are slim. For simplicity, let us suppose further [for instance, by replacing X by a finite étale covering of X] that g X 2. Then there exists a functorial “group-theoretic” algorithm for reconstructing the function field 46 SHINICHI MOCHIZUKI K X from the extension of profinite groups 1 Δ η X Π η X G k 1; this algorithm consists of the following steps: (a) Let l be a prime number. If ρ : G k Z × l is a character, and M is an abelian pro-l group equipped with a continuous action by G k , then let us write F ρ (M ) M for the closed subgroup topologically generated by the closed subgroups of M that are isomorphic to Z l (ρ) [i.e., the G k - module obtained by letting G k act on Z l via ρ] as H-modules, for some open subgroup H G k . [Thus, F ρ (M ) M depends only on the “power- equivalence class” of ρ cf. Remark 1.5.1.] Then the power-equivalence class of the cyclotomic character χ l : G k Z × l may be characterized Z ) is not topologically finitely generated. by the condition that F χ l ab l η X (b) Let l be a prime number. If M is an abelian pro-l group equipped with a continuous action by G k such that M/F χ l (M ) is topologically finitely generated, then let us write M  T (M ) for the maximal torsion- free quasi-trivial quotient [i.e., maximal torsion-free quotient on which G k acts through a finite quotient]. [Thus, one verifies immediately that “T (M )” is well-defined.] Then one may compute the genus of X via the formula [cf. the proof of [Mzk21], Corollary 2.10] ab 2g X = dim Q l (Q(Δ ab η X Z l ) Q l ) + dim Q l (T η X Z l ) Q l ) def where we write Q(−) = (−)/F χ l (−). In particular, this allows one to characterize, via the Hurwitz formula, those pairs of open subgroups J i H i Δ η X such that “the covering between J i and H i is cyclic of or- der a power of l and totally ramified at precisely one closed point but un- ramified elsewhere” [cf. the proof of [Mzk21], Corollary 2.10]. Moreover, this last characterization implies a “group-theoretic” characterization of the inertia subgroups I x Δ η X of points x X(k) [cf. the proof of [Mzk21], Corollary 2.10; the latter portion of the proof of [Mzk9], Lemma 1.3.9], hence of the quotient Δ η X  Δ X [whose kernel is topologically normally generated by the I x , for x X(k)]. Finally, the decomposition group D x Π η X of x X(k) may then be constructed as the normal- izer [or, equivalently, commensurator] of I x in Π η X [cf., e.g., [Mzk12], Theorem 1.3, (ii)]. (c) One may construct the natural isomorphisms I x M X [where x X(k); M X is as in Proposition 1.4, (ii)] via the technique of Proposition 1.4, (ii). These isomorphisms determine [by restriction to the I x ] a natural map   Z H 1 η X , M X ) x∈X(k) [cf. Proposition 1.6, (iii)]. Denote by P η X H 1 η X , M X ) [cf. Proposi- tion 1.8] the inverse image in H 1 η X , M X ) of the subgroup of  x∈X(k)  Z TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 47 consisting of the principal divisors i.e., divisors D of degree zero supported on a collection of points X(k H ), where k H k is the subfield corresponding to an open subgroup H G k , whose associated class H 1 (H, Δ ab X ) [i.e., the class obtained as the difference between the section “t D of Proposition 1.6, (ii), and the identity section] is trivial. (d) The image of the Kummer map × K X H 1 η X , M X ) may be constructed as the subgroup generated by elements θ P η X for which there exists an x X(k) such that θ| x H 1 (D x , M X ) vanishes [i.e., = 1, if one works multiplicatively] cf. the technique of Proposition × {0} may be recovered 1.8, (i). Moreover, the additive structure on K X via the algorithm of Proposition 1.3. Finally, the asserted “functoriality” is with respect to arbitrary open injective homomorphisms of extensions of profinite groups [cf. Remark 1.10.1, (i)]. Proof. The slimness of Δ η X follows immediately from the argument applied to verify the slimness portion of [Mzk21], Corollary 2.10. The validity of the recon- struction algorithm asserted in Theorem 1.11 is immediate from the various results cited in the statement of this algorithm. Now, by applying the functoriality of this algorithm, the slimness of G k follows immediately from the argument applied in [Mzk5], Lemma 15.8, to verify the slimness of G k when k is sub-p-adic. Finally, the slimness of Π η X follows from the slimness of Δ η X , G k . Remark 1.11.1. (i) One verifies immediately that when k is an MLF, the semi-absolute algo- rithms of Theorem 1.11 may be rendered absolute [i.e., one may construct the kernel of the quotient “Π η X  G k ”] by applying the algorithm that is implicit in the proof of the corresponding portion of [Mzk21], Corollary 2.10. (ii) Suppose, in the notation of Theorem 1.11 that k is an NF. Then an absolute version of the functoriality portion [i.e., the “Grothendieck Conjecture” portion] of Theorem 1.11 is proven in [Pop] [cf. [Pop], Theorem 2]. Moreover, in [Pop], Observation [and the following discussion], an algorithm is given for passing from the absolute data “Π η X to the semi-absolute data “(Π η X , Δ η X Π η X )”. Thus, by combining this algorithm of [Pop] with Theorem 1.11, one obtains an absolute version of Theorem 1.11. Remark 1.11.2. One may think of the argument used to prove the slimness of G k in the proof of Theorem 1.11 [i.e., the argument of the proof of [Mzk5], Lemma 15.8] as being similar in spirit to the proof [cf., e.g., [Mzk9], Theorem 1.1.1, (ii)] of the slimness of G k via local class field theory in the case where k is an MLF, as well as to the proof of the slimness of the geometric fundamental group of a hyperbolic curve given, for instance, in [MT], Proposition 1.4, via the induced action on the 48 SHINICHI MOCHIZUKI torsion points of the Jacobian of the curve, in which the curve may be embedded. That is to say, in the case where k is an arbitrary Kummer-faithful field, since one does not have an analogue of local class field theory (respectively, of the embedding of a curve in its Jacobian), the moduli of hyperbolic curves over k, in the context of a relative anabelian result for the arithmetic fundamental groups of such curves, plays the role of the abelianization of G k (respectively, of the torsion points of the Jacobian) in the case where k is an MLF (respectively, in the case of the geometric fundamental group of a hyperbolic curve) i.e., the role of a “functorial, group- theoretically reconstructible embedding” of k (respectively, the curve). Remark 1.11.3. It is interesting to note that the techniques that appear in the algorithms of Theorem 1.11 are extremely elementary. For instance, unlike the case with Theorem 1.9, Corollary 1.10, the algorithms of Theorem 1.11 do not depend on the somewhat difficult [e.g., in their use of p-adic Hodge theory] results of [Mzk5]. Put another way, this elementary nature of Theorem 1.11 serves to highlight the fact that the only non-elementary portion [in the sense of its dependence of the results of [Mzk5]] of the algorithms of Theorem 1.9 is the use of the technique of Belyi cuspidalizations. It is precisely this “non-elementary portion” of Theorem 1.9 that requires us, in Theorem 1.9, to assume that the base field is sub-p-adic [i.e., as opposed to merely Kummer-faithful, as in Theorem 1.11]. Remark 1.11.4. (i) The observation of Remark 1.11.3 prompts the following question: If the birational version [i.e., Theorem 1.11] of Theorem 1.9 is so much more elementary than Theorem 1.9, then what is the advantage [i.e., relative to the anabelian geometry of function fields] of considering the anabelian geometry of hyperbolic curves? One key advantage of working with hyperbolic curves, in the context of the theory of the present paper, lies in the fact that “most” hyperbolic curves admit a core [cf. [Mzk3], §3; [Mzk10], §2]. Moreover, the existence of “cores” at the level of schemes has a tendency to imply to existence of “cores” at the level of “étale fundamental groups considered geometrically”, i.e., at the level of anabelioids [cf. [Mzk11], §3.1]. The existence of a core is crucial to, for instance, the theory of the étale theta function given in [Mzk18], §1, §2, and, moreover, in the present three-part series, plays an important role in the theory of elliptically admissible [cf. [Mzk21], Definition 3.1] hyperbolic orbicurves. On the other hand, it is easy to see that “function fields do not admit cores”: i.e., if, in the notation of Theorem 1.11, we write Loc(η X ) for the category whose objects are connected schemes that admit a connected finite étale covering which is also a connected finite étale covering of η X , and whose morphisms are the finite étale morphisms, then Loc(η X ) fails to admit a terminal object. (ii) The observation of (i) is interesting in the context of the theory of §5 below, in which we apply various [mono-]anabelian results to construct “canonical rigid integral structures” called “log-shells”. Indeed, in the Introduction to TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 49 [Mzk11], it is explained, via analogy to the complex analytic theory of the upper half-plane, how the notion of a core may be thought of as a sort of “canonical integral structure” i.e., relative to the “modifications of integral structure” constituted by “going up and down via various finite étale coverings”. Here, it is interesting to note that this idea of a “canonical integral structure relative to going up and down via finite étale coverings” may also be seen in the theory surrounding the property of cyclotomic rigidity in the context of the étale theta function [cf., e.g., [Mzk18], Remark 2.19.3]. Moreover, let us observe that these “integral structures with respect to finite étale coverings” may be thought of as “exponentiated integral structures” i.e., in the sense that, for instance, in the case of G m over Q, these integral structures are not integral structures relative to the scheme-theoretic base ring Z Q, but rather with respect to the exponent of the standard coordinate U , which, via multiplication by various nonnegative integers N , gives rise, in the form of mappings U n → U N ·n , to various finite étale coverings of G m . Such “non-scheme- theoretic exponentiated copies of Z” play an important role in the theory of the étale theta function as the Galois group of a certain natural infinite étale covering of the Tate curve cf. the discussion of [Mzk18], Remark 2.16.2. Moreover, the idea of constructing “canonical integral structures” by “de-exponentiating certain exponentiated integral structures” may be rephrased as the idea of “constructing canonical integral structures by applying some sort of logarithm operation”. From this point of view, such “canonical integral structures with respect to finite étale coverings” are quite reminiscent of the canonical integral structures arising from log-shells to be constructed in §5 below. Remark 1.11.5. In the context of the discussion of Remark 1.11.4, if the hyper- bolic curve in question is affine, then, relative to the function field of the curve, the additional data necessary to determine the given affine hyperbolic curve consists precisely of some [nonempty] finite collection of conjugacy classes of inertia groups [i.e., “I x as in Theorem 11.1, (b)]. Thus, from the point of view of the discussion of Remark 1.11.3, the technique of Belyi cuspidalizations is applied precisely so as to enable one to work with this additional data [cf. also the discussion of Remark 3.7.7 below]. 50 SHINICHI MOCHIZUKI Section 2: Archimedean Reconstruction Algorithms In the present §2, we re-examine various aspects of the complex analytic the- ory of [Mzk14] from an algorithm-based, “model-implicit” [cf. Remark 2.7.4 below] point of view motivated by the Galois-theoretic theory of §1. More precisely, the “SL 2 (R)-based approach” of [Mzk14], §1, may be seen in the general theory of Aut-holomorphic spaces given in Proposition 2.2, Corollary 2.3, while the “paral- lelograms, rectangles, squares approach” of [Mzk14], §2, is developed further in the reconstruction algorithms of Propositions 2.5, 2.6. These two approaches are com- bined to obtain the main result of the present §2 [cf. Corollary 2.7], which consists of a certain reconstruction algorithm for the “local linear holomorphic structure” of an Aut-holomorphic orbispace arising from an elliptically admissible hyperbolic orbicurve. Finally, in Corollaries 2.8, 2.9, we consider the relationship between Corollary 2.7 and the global portion of the Galois-theoretic theory of §1. The following definition will play an important role in the theory of the present §2. Definition 2.1. (i) Let X be a Riemann surface [i.e., a complex manifold of dimension one]. Write A X for the assignment that assigns to each connected open subset U X the group def A X (U ) = Aut hol (U ) of holomorphic automorphisms of U which we think of as being “some distin- guished subgroup” of the group of self-homeomorphisms Aut(U top ) of the under- lying topological space U top of U . We shall refer to as the Aut-holomorphic space associated to X the pair def X = (X top , A X ) def consisting of the underlying topological space X top = X top of X, together with the def assignment A X = A X ; also, we shall refer to the assignment A X = A X as the Aut- holomorphic structure on X top = X top [determined by X]. If X is biholomorphic to the open unit disc, then we shall refer to X as an Aut-holomorphic disc. If X is a hyperbolic Riemann surface of finite type (respectively, a hyperbolic Riemann surface of finite type associated to an elliptically admissible [cf. [Mzk21], Definition 3.1] hyperbolic curve over C), then we shall refer to the Aut-holomorphic space X as hyperbolic of finite type (respectively, elliptically admissible). If U is a collection of connected open subsets of X that forms a basis for the topology of X and, moreover, satisfies the condition that any connected open subset of X that is contained in an element of U is itself an element of U , then we shall refer to U as a local structure on X top and to the restriction A X | U of A X to U as a [U-local] pre-Aut-holomorphic structure on X top . (ii) Let X (respectively, Y ) be a Riemann surface; X (respectively, Y) the Aut-holomorphic space associated to X (respectively, Y ); U (respectively, V) a local structure on X top (respectively, Y top ). Then we shall refer to as a (U, V)-local morphism of Aut-holomorphic spaces φ : X Y TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 51 any local isomorphism of topological spaces φ top : X top Y top with the property that for any open subset U X U that maps homeomorphically via φ top onto some open subset U Y V, φ top induces a bijection A X (U X ) A Y (U Y ); when U, V are, respectively, the sets of all connected open subsets of X, Y , then we shall omit the word “(U , V)-local” from this terminology; when φ top is a finite covering space map, we shall say that φ is finite étale. We shall refer to a map X Y which is either holomorphic or anti-holomorphic at each point of X as an RC-holomorphic morphism [cf. [Mzk14], Definition 1.1, (vi)]. (iii) Let Z, Z  be orientable topological surfaces [i.e., two-manifolds]. If p Z, then let us write def ab Orn(Z, p) = lim −→ π 1 (W \{p}) W where W ranges over the connected open neighborhoods of p in Z; “π 1 (−)” denotes the usual topological fundamental group, relative to some basepoint [so “π 1 (−)” is only defined up to inner automorphisms, an indeterminacy which may be eliminated by passing to the abelianization “ab”]; thus, Orn(Z, p) is [noncanon- ically!] isomorphic to Z. Note that since Z is orientable, it follows that the assign- ment p → Orn(Z, p) determines a trivial local system on Z, whose module of global sections we shall denote by Orn(Z) [so Orn(Z) is a direct product of copies of Z, in- dexed by the connected components of Z]. One verifies immediately that any local isomorphism Z Z  induces a well-defined homomorphism Orn(Z) Orn(Z  ). We shall say that any two local isomorphisms α, β : Z Z  are co-oriented if they induce the same homomorphism Orn(Z) Orn(Z  ). We shall refer to as a pre-co-orientation ζ : Z Z  any equivalence class of local isomorphisms Z Z  relative to the equivalence relation determined by the property of being co-oriented [so a pre-co-orientation may be thought of as a collection of maps Z Z  , or, al- ternatively, as a homomorphism Orn(Z) Orn(Z  )]. Thus, the pre-co-orientations from the open subsets of Z to Z  form a pre-sheaf on Z; we shall refer to as a co-orientation ζ : Z Z  any section of the sheafification of this pre-sheaf [so a co-orientation may be thought of as a collection of maps from open subsets of Z to Z  , or, alternatively, as a homomorphism Orn(Z) Orn(Z  )]. (iv) Let X, Y , X, Y, U, V be as in (ii). Then we shall say that two (U, V)-local morphisms of Aut-holomorphic spaces φ 1 , φ 2 : X Y are co-holomorphic if φ top 1 and φ top are co-oriented [cf. (iii)]. We shall refer to as a pre-co-holomorphicization 2 ζ : X Y any equivalence class of (U, V)-local morphisms of Aut-holomorphic spaces X Y relative to the equivalence relation determined by the property of being co-holomorphic [so a pre-co-holomorphicization may be thought of as a collection of maps from X top to Y top ]. Thus, the pre-co-holomorphicizations from the Aut-holomorphic spaces determined by open subsets of X top to Y form a pre- sheaf on X top ; we shall refer to as a co-holomorphicization [cf. also Remark 2.3.2 below] ζ : X Y any section of the sheafification of this pre-sheaf [so a co-holomorphicization may be thought of as a collection of maps from open subsets of X top to Y top ]. Finally, we 52 SHINICHI MOCHIZUKI observe that every co-holomorphicization (respectively, pre-co-holomorphicization) determines a co-orientation (pre-co-orientation) between the underlying topological spaces. Remark 2.1.1. One verifies immediately that there is a natural extension of the notions of Definition 2.1 to the case of Riemann orbisurfaces, which give rise to “Aut-holomorphic orbispaces” [not to be confused with the “orbi-objects” of §0, which will always be identifiable in the present paper by means of the hyphen “-” following the prefix “orbi”]. Here, we understand the term “Riemann orbisurface” to refer to a one-dimensional complex analytic stack which is locally isomorphic to the complex analytic stack obtained by forming the stack-theoretic quotient of a Riemann surface [i.e., a one-dimensional complex manifold] by a finite group of [holomorphic] automorphisms [of the Riemann surface]. In particular, a “Riemann orbiface” is necessarily a Riemann surface over the complement, in the “coarse space” associated to the orbisurface, of some discrete closed subset. Remark 2.1.2. One important aspect of the “Aut-holomorphic” approach to the notion of a “holomorphic structure” is that this approach has the virtue of being free of any mention of some “fixed reference model” copy of the field of complex numbers C cf. Remark 2.7.4 below. Proposition 2.2. (Commensurable Terminality of RC-Holomorphic Automorphisms of the Disc) Let X, Y be Aut-holomorphic discs, arising, respectively, from Riemann surfaces X, Y . Then: (i) Every isomorphism of Aut-holomorphic spaces X Y arises from a unique RC-holomorphic isomorphism X Y . (ii) Let us regard the group Aut(X top ) as equipped with the compact-open topology. Then the subgroup Aut RC-hol (X) Aut(X top ) of RC-holomorphic automorphisms of X, which [as is well-known] contains Aut hol (X) as a subgroup of index two, is closed and commensurably terminal [cf. [Mzk20], §0]. Moreover, we have isomorphisms of topological groups Aut hol (X) = SL 2 (R)/{±1}; Aut RC-hol (X) = GL 2 (R)/R × [where we regard Aut hol (X), Aut RC-hol (X), as equipped with the topology induced by the topology of Aut(X top ), i.e., the compact-open topology]. Proof. It is immediate from the definitions that assertion (i) follows formally from the commensurable terminality [in fact, in this situation, normal terminality suffices] of assertion (ii). Thus, it suffices to verify assertion (ii). First, we recall that we have a natural isomorphism of connected topological groups Aut hol (X) = SL 2 (R)/{±1} TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 53 [where we regard Aut hol (X) as equipped with the compact-open topology]. Next, let us recall the well-known fact in elementary complex analysis that “a sequence of holomorphic functions on X top that converges uniformly on compact subsets of X top converges to a holomorphic function on X top ”. [This fact is often ap- plied in proofs of the Riemann mapping theorem.] This fact implies immedi- ately that Aut hol (X), Aut RC-hol (X) are closed in Aut(X top ). Now suppose that α Aut(X top ) lies in the commensurator of Aut hol (X); thus, the intersection · Aut hol (X)·α −1 ) Aut hol (X) is a closed subgroup of finite index of Aut hol (X). But this implies that (α·Aut hol (X)·α −1 ) Aut hol (X) is an open subgroup of Aut hol (X), hence [since Aut hol (X) is connected] that · Aut hol (X) · α −1 ) Aut hol (X) = Aut hol (X), i.e., that α · Aut hol (X) · α −1 Aut hol (X). Thus, by replacing α by α −1 , we conclude that α normalizes Aut hol (X), i.e., that α induces an automor- phism of the topological group Aut hol (X) = SL 2 (R)/{±1}, hence also [by Cartan’s theorem cf., e.g., [Serre], Chapter V, §9, Theorem 2; the proof of [Mzk14], Lemma 1.10] of the real analytic Lie group SL 2 (R)/{±1}. Thus, as is well-known, it fol- lows [for instance, by considering the action of α on the Borel subalgebras of the complexification of the Lie algebra of SL 2 (R)/{±1}] that α arises from an element of GL 2 (C)/C × that fixes [relative to the action by conjugation] the Lie subalgebra sl 2 (R) of sl 2 (C). But such an element of GL 2 (C)/C × is easily verified to be an element of GL 2 (R)/R × . In particular, by considering the action of α on maximal compact subgroups of Aut hol (X) [cf. the proof of [Mzk14], Lemma 1.10], it follows that α arises from an RC-holomorphic automorphism of X, as desired. In fact, as the following result shows, the notions of an Aut-holomorphic struc- ture and a pre-Aut-holomorphic structure are equivalent to one another, as well as to the usual notion of a “holomorphic structure”. Corollary 2.3. (Morphisms of Aut-Holomorphic Spaces) Let X (respec- tively, Y ) be a Riemann surface; X (respectively, Y) the Aut-holomorphic space associated to X (respectively, Y ); U (respectively, V) a local structure on X top (re- spectively, Y top ). Then: (i) Every (U , V)-local morphism of Aut-holomorphic spaces φ : X Y arises from a unique étale RC-holomorphic morphism ψ : X Y . Moreover, if, in this situation, X, Y [i.e., X top , Y top ] are connected, then there exist pre- cisely two co-holomorphicizations X Y, corresponding to the holomorphic and anti-holomorphic local isomorphisms from open subsets of X to Y . (ii) Every pre-Aut-holomorphic structure on X top extends to a unique Aut-holomorphic structure on X top . Proof. Assertion (i) follows immediately from the definitions, by applying Propo- sition 2.2, (i), to sufficiently small open discs in X top . Assertion (ii) follows im- mediately from assertion (i) by applying assertion (i) to automorphisms of the Aut-holomorphic spaces determined by arbitrary connected open subsets of X top which determine the same co-holomorphicization as the identity automorphism. 54 SHINICHI MOCHIZUKI Remark 2.3.1. Note that Corollary 2.3 may be thought of as one sort of “com- plex analytic analogue of the Grothendieck Conjecture”, that, although formulated somewhat differently, contains [to a substantial extent] the same essential mathe- matical content as [Mzk14], Theorem 1.12 cf. the similarity between the proofs of Proposition 2.2 and [Mzk14], Lemma 1.10; the application of the p-adic version of Cartan’s theorem in the proof of [Mzk8], Theorem 1.1 [i.e., in the proof of [Mzk8], Lemma 1.3]. Remark 2.3.2. It follows, in particular, from Corollary 2.3, (ii), that [in the notation of Definition 2.1, (iv)] the notion of a co-holomorphicization X Y is, in fact, independent of the choice of the local structures U , V. Remark 2.3.3. It follows immediately from Corollary 2.3, (i), that any composite of morphisms of Aut-holomorphic spaces is again a morphism of Aut-holomorphic spaces. Corollary 2.4. (Holomorphic Arithmeticity and Cores) Let X be a hyper- bolic Aut-holomorphic space of finite type associated to a Riemann surface X [which is, in turn, determined by a hyperbolic curve over C]. Then one may determine the arithmeticity [in the sense of [Mzk3], §2] of X and, when X is not arithmetic, construct the Aut-holomorphic orbispace [cf. Remark 2.1.1] associated to the hyperbolic core [cf. [Mzk3], Definition 3.1] of X, via the following func- torial algorithm, which involves only the Aut-holomorphic space X as input data: (a) Let U top X top be any universal covering of X top [i.e., a connected covering space of the topological space X top which does not admit any non- trivial connected covering spaces]. Then one may construct the funda- mental group π 1 (X top ) as the group of automorphisms Aut(U top /X top ) of U top over X top . (b) In the notation of (a), by considering the local structure on U top consisting of connected open subsets of U top that map isomorphically onto open subsets of X top , one may construct a natural pre-Aut-holomorphic structure on U top hence also [cf. Corollary 2.3, (ii)] a natural Aut- holomorphic structure on U top by restricting the Aut-holomorphic structure of X on X top ; denote the resulting Aut-holomorphic space by U. Thus, we obtain a natural injection π 1 (X top ) = Aut(U top /X top ) → Aut 0 (U) Aut(U) where we recall [cf. Proposition 2.2, (ii); Corollary 2.3, (i)] that Aut(U), equipped with the compact-open topology, is isomorphic, as a topological group, to GL 2 (R)/R × ; we write Aut 0 (U) Aut(U) for the connected component of the identity of Aut(U). (c) In the notation of (b), X is not arithmetic if and only if the image of π 1 (X top ) in Aut 0 (U) is of finite index in its commensurator Π TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 55 Aut 0 (U) in Aut 0 (U) [cf. [Mzk3], §2, §3]. If X is not arithmetic, then the Aut-holomorphic orbispace X H associated to the hyperbolic core H of X may be constructed by form- ing the “orbispace quotient” of U top by Π and equipping this quotient with the pre-Aut-holomorphic structure which [cf. Corollary 2.3, (ii)] deter- mines a unique Aut-holomorphic structure determined by restrict- ing the Aut-holomorphic structure of U to some suitable local structure as in (b). Finally, the asserted “functoriality” is with respect to finite étale morphisms of Aut-holomorphic spaces arising from hyperbolic curves over C. Proof. The validity of the algorithm asserted in Corollary 2.4 is immediate from the constructions that appear in the statement of this algorithm [together with the references quoted in these constructions]. Remark 2.4.1. One verifies immediately that Corollary 2.4 admits a natural extension to the case where X arises from a hyperbolic orbicurve over C [cf. Remark 2.1.1]. Remark 2.4.2. Relative to the analogy with the theory of §1 [cf. Remark 2.7.3 below], Corollary 2.4 may be regarded as a sort of holomorphic analogue of results such as [Mzk10], Theorem 2.4, concerning categories of finite étale localizations of hyperbolic orbicurves. Next, we turn our attention to re-examining, from an algorithm-based point of view, the theory of affine linear structures on Riemann surfaces in the style of [Mzk14], §2; [Mzk14], Appendix. Following the terminology of [Mzk14], Definition A.3, (i), (ii), we shall refer to as “parallelograms”, “rectangles”, or “squares” the distinguished open subsets of C = R + iR which are of the form suggested by these respective terms. Proposition 2.5. Squares) Let (Linear Structures via Parallelograms, Rectangles, or U C = R + iR be a connected open subset. Write S(U ) R(U ) P(U ) for the sets of pre-compact squares, rectangles, and parallelograms in U ; let Q {S, R, P}. Then there exists a functorial algorithm for constructing the parallel line segments, parallelograms, orientations, and “local additive structures” [in the sense described below] of U that involves only the input data (U, Q(U )) i.e., consisting of the abstract set U , equipped with the datum of a 56 SHINICHI MOCHIZUKI collection of distinguished open subsets Q(U ) [which clearly forms a basis for, hence determines, the topology of U ] as follows: (a) Define a strict line segment L of U to be an intersection of the form def L = Q 1 Q 2 where Q 1 , Q 2 are the respective closures of Q 1 , Q 2 Q(U ); Q 1 Q 2 = ∅; L is of infinite cardinality. Define two strict line segments to be strictly collinear if their intersection is of infinite cardinality. Define a strict chain of U to be a finite ordered set of strict line segments L 1 , . . . , L n [where n 2 is an integer] such that L i , L i+1 are strictly collinear for i = 1, . . . , n 1. Then one constructs the [closed, bounded] line segments of U by observing that a line segment may be character- ized as the union of strict line segments contained a strict chain of U ; an endpoint of a line segment L is a point of the boundary ∂L of L [i.e., a point whose complement in L is connected]. (b) Define a ∂Q-parallelogram of U to be a closed subset of U of the form def ∂Q = Q\Q where Q Q(U ); Q denotes the closure of Q. Define a side of a parallelogram Q Q(U ) to be a maximal line segment contained in the ∂Q-parallelogram ∂Q. Define two line segments L, L  of U to be strictly parallel if there exist non-intersecting sides S, S  of a parallel- ogram Q(U ) such that S L, S  L  . Then one constructs the pairs (L, L  ) of parallel line segments by observing that L, L  are parallel if and only if L is equivalent to L  relative to the equivalence relation on line segments generated by the relation of inclusion and the relation of being strictly parallel. (c) Define a pre-∂-parallelogram ∂P of U to be a union of the mem- bers of a family of four line segments {L i } i∈Z/4Z of U such that for any two distinct points p 1 , p 2 ∂P , there exists a line segment L such that ∂L = {p 1 , p 2 }, and, moreover, for each i Z/4Z, L i and L i+2 are paral- lel and non-intersecting, and we have an equality of sets L i L i+1 = (∂L i ) (∂L i+1 ) of cardinality one. If ∂P is a pre-∂-parallelogram of U , then define the associated pre-parallelogram of U to be the union of line segments L of U such ∂L ∂P . Then one constructs the parallelograms P(U ) of U as the interiors of the pre-parallelograms of U . (d) Let p U . Define a frame F = (S 1 , S 2 ) of U at p to be an ordered pair of distinct intersecting sides S 1 , S 2 of a parallelogram P P(U ) such that S 1 S 2 = {p}; in this situation, we shall refer to any line segment of U that has infinite intersection with P as being framed by F . Define two frames F = (S 1 , S 2 ), F  = (S 1  , S 2  ) of U at p to be strictly co-oriented if S 1  is framed by F , and S 2 is framed by F  . Then one constructs the orientations of U at p [of which there are precisely 2] by observing that an orientation of U at p may be characterized as an equivalence class of frames of U at p, relative to the equivalence relation on frames of U at p generated by the relation of being strictly co-oriented. TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 57 (e) Let p U . Then given a, b U , the sum a + p b U , relative to the origin p i.e., the “local additive structure” of U at p may be constructed, whenever it is defined, in the following fashion: If a = p, then a + p b = b; if b = p, then a + p b = a. If a, b  = p, then for P P(U ) such that P contains [distinct] intersecting sides S a , S b for which S a S b = {p}, ∂S a = {p, a}, ∂S b = {p, b}, we take a + p b to be the unique endpoint of a side of P that ∈ {a, b, p}. [Thus, “a + p b” is defined for a, b in some neighborhood of p in U .] Finally, the asserted “functoriality” is with respect to open immersions [of ab- stract topological spaces] ι : U 1 → U 2 [where U 1 , U 2 C are connected open subsets] such that ι maps Q(U 1 ) into Q(U 2 ). Proof. The validity of the algorithm asserted in Proposition 2.5 is immediate from the elementary content of the characterizations contained in the statement of this algorithm. Remark 2.5.1. We shall refer to a frame of U at p U as orthogonal if it arises from an ordered pair of distinct intersecting sides of a rectangle R(U ) P(U ). Proposition 2.6. (Local Linear Holomorphic Structures via Rectangles or Squares) Let U , S, R, P, Q be as in Proposition 2.5; suppose further that Q  = P. Then there exists a functorial algorithm for constructing the “local linear holomorphic structure” [in the sense described below] of U that involves only the input data (U, Q(U )) i.e., consisting of the abstract topological space U , equipped with the datum of a collection of distinguished open subsets Q(U ) as follows: (a) For p U , write A p for the group of automorphisms of the projective system of connected open neighborhoods of p in U that are compatible with the “local addi- tive structures” of Proposition 2.5, (e), and preserve the orthogonal frames and orientations [at p] of Proposition 2.5, (d); Remark 2.5.1. Also, we equip A p with the topology induced by the topologies of the open neighborhoods of p that A p acts on; note that the “local additive struc- tures” of Proposition 2.5, (e), determine an additive structure, hence also a topological field structure on A p {0}. Then we have a natural isomorphism of topological groups C × A p [induced by the tautological action of C × on C U ] that is compatible with the topological field structures on the union of either side with “{0}”. In particular, one may construct “C × at p” i.e., the “local linear holomorphic structure” of U at p by thinking of this “lo- cal linear holomorphic structure” as being constituted by the topological 58 SHINICHI MOCHIZUKI field A p {0}, equipped with its tautological action on the projective system of open neighborhoods of p. (b) For p, p  U , one constructs a natural isomorphism of topological groups A p A p  that is compatible with the topological field structures on either side as follows: If p  is sufficiently close to p, then the “local additive struc- tures” of Proposition 2.5, (e), determine homeomorphisms [by “trans- lation”, i.e., “addition”] from sufficiently small neighborhoods of p onto sufficiently small neighborhoods of p  ; these homeomorphisms thus induce the desired isomorphism A p A p  . Now, by joining an arbitrary p  to p via a chain of “sufficiently small open neighborhoods” and composing the resulting isomorphisms of “local linear holomorphic structures”, one ob- tains the desired isomorphism A p A p  for arbitrary p, p  U . Finally, this isomorphism is independent of the choice of a chain of “sufficiently small open neighborhoods” used in its construction. Finally, the asserted “functoriality” is to be understood in the same sense as in Proposition 2.5. Proof. The validity of the algorithm asserted in Proposition 2.6 is immediate from the elementary content of the characterizations contained in the statement of this algorithm. Remark 2.6.1. Thus, the algorithms of Propositions 2.5, 2.6 may be regarded as superseding the techniques applied in the proof of [Mzk14], Proposition A.4. Moreover, just as the theory of [Mzk14], Appendix, was applied in [Mzk14], §2, one may apply the algorithms of Propositions 2.5, 2.6 to give algorithms for recon- structing the local linear and orthogonal structures on a Riemann surface equipped with a nonzero square differential from the various categories which are the topic of [Mzk14], Theorem 2.3. We leave the routine details to the interested reader. Corollary 2.7. (Local Linear Holomorphic Structures via Holomorphic Elliptic Cuspidalization) Let X be an elliptically admissible Aut-holomorphic orbispace [cf. Remark 2.1.1] associated to a Riemann orbisurface X. Then there exists a functorial algorithm for constructing the “local linear holo- morphic structure” [cf. Proposition 2.6] on X top that involves only the Aut- holomorphic space X as input data, as follows: (a) By the definition of “elliptically admissible”, we may apply Corollary 2.4, (c), to construct the [Aut-holomorphic orbispace associated to the] semi-elliptic hyperbolic core X H of X [i.e., X], together with the unique [cf. [Mzk21], Remark 3.1.1] double covering E H by an Aut- holomorphic space [i.e., the covering determined by the unique torsion- free subgroup of index two of the group Π of Corollary 2.4, (c)]. [Thus, TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III E is the Aut-holomorphic space associated to a once-punctured elliptic curve.] (b) By considering “elliptic cuspidalization diagrams” as in [Mzk21], Example 3.2 [cf. also the equivalence of Corollary 2.3, (i)] E ← U E where U E is an abelian finite étale covering [which necessarily extends to a covering of the one-point compactification of E top ]; E top ← U top is an open immersion whose image is the complement of a finite subset of E top ; E ← U, U E are co-holomorphic one may construct the torsion points of [the elliptic curve determined by] E as the points in the complement of the image of such morphisms U → E, together with the group structure on these torsion points [which is induced by the group structure of the Galois group Gal(U/E)]. (c) Since the torsion points of (b) are dense in E top , one may construct the group structure on [the one-point compactification of ] E top [that arises from the elliptic curve determined by E] as the unique topolog- ical group structure that extends the group structure on the torsion points of (b). This group structure determines “local additive struc- tures” [cf. Proposition 2.5, (e)] at the various points of E top . Moreover, by considering one-parameter subgroups of these local additive group struc- tures, one constructs the line segments [cf. Proposition 2.5, (a)] of E top ; by considering translations of line segments, relative to these local additive group structures, one constructs the pairs of parallel line segments [cf. Proposition 2.5, (b)] of E top , hence also the parallelograms, frames, and orientations [cf. Proposition 2.5, (c), (d)] of E top . (d) Let V be the Aut-holomorphic space determined by a parallelogram V top E top [cf. (c)]. Then the one-parameter subgroups of the [topological] group A V (V top ) [ = SL 2 (R)/{±1} cf. Proposition 2.2, (ii); Corollary 2.3, (i); the Riemann mapping theorem of elementary complex analysis] are precisely the closed connected subgroups for which the complement of some connected open neighborhood of the identity element fails to be connected. If S is a one-parameter subgroup of A V (V top ), p V top , and L is a line segment one of whose endpoints is equal to p, then L is tangent to S · p at p if and only if any pairs of sequences of points of L\{p}, (S · p)\{p}, converge to the same element of the quotient space V top \{p}  P(V, p) determined by identifying positive real multiples of elements of V top \{p}, relative to the local additive structure at p. In particular, one may con- struct the orthogonal frames of E top as the frames consisting of pairs of line segments L 1 , L 2 emanating from a point p E top that are tan- gent, respectively, to orbits S 1 · p, S 2 · p of one-parameter subgroups S 1 , S 2 A V (V top ) such that S 2 is obtained from S 1 by conjugating S 1 by an element of order four [i.e., “±i”] of a compact one-parameter subgroup [i.e., a “one-dimensional torus”] of A V (V top ) that fixes p. 59 60 SHINICHI MOCHIZUKI (e) For p E top , write A p for the group of automorphisms of the projective system of connected open neighborhoods of p in E top that are compatible with the “local ad- ditive structures” of (c) and preserve the orthogonal frames and orientations [at p] of (c), (d) [cf. Proposition 2.6, (a)]. Then just as in Proposition 2.6, (a), we obtain topological field structures on A p {0}, together with compatible isomorphisms A p A p  , for p  E top . This system of “A p ’s” may be thought of as a system of “local linear holomorphic structures” on E top or X top . Finally, the asserted “functoriality” is with respect to finite étale morphisms of Aut-holomorphic orbispaces arising from hyperbolic orbicurves over C. Proof. The validity of the algorithm asserted in Corollary 2.7 is immediate from the constructions that appear in the statement of this algorithm [together with the references quoted in these constructions]. Remark 2.7.1. It is by no means the intention of the author to assert that the technique applied in Corollary 2.7, (b), (c), to recover the “local additive structure” via elliptic cuspidalization is the unique way to construct this local additive struc- ture. Indeed, perhaps the most direct approach to the problem of constructing the local additive structure is to compactify the given once-punctured elliptic curve and then to consider the group structure of the [connected component of the identity of the] holomorphic automorphism group of the resulting elliptic curve. By com- parison to this direct approach, however, the technique of elliptic cuspidalization has the virtue of being compatible with the “hyperbolic structure” of the hyperbolic orbicurves involved. In particular, it is compatible with the various “hyperbolic fun- damental groups” of these orbicurves. This sort of compatibility with fundamental groups plays an essential role in the nonarchimedean theory [cf., e.g., the theory of [Mzk18], §1, §2]. On the other hand, the “direct approach” described above is not entirely unrelated to the approach via elliptic cuspidalization in the sense that, if one thinks of the torsion points in the latter approach as playing an analogous role to the role played by the “entire compactified elliptic curve” in the former approach, then the latter approach may be thought of as a sort of discretization via torsion points cf. the point of view of Hodge-Arakelov theory, as discussed in [Mzk6], [Mzk7] of the former approach. Here, we note that the density of torsion points in the archimedean theory of the elliptic cuspidalization is reminiscent of the density of NF-points in the nonarchimedean theory of the Belyi cuspidalization [cf. §1]. Remark 2.7.2. In light of the role played by the technique of elliptic cuspidal- ization both in Corollary 2.7 and in the theory of [Mzk18], §1, §2, it is of interest to compare these two theories. From an archimedean point of view, the theory of [Mzk18] may be roughly summarized as follows: One begins with the uniformization G  G/q Z E TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 61 of an elliptic curve E over C by a copy G of C × . Here, the “q-parameter” of E may be thought of as being an element def q H = G Gal(G/E) [where we recall that Gal(G/E) = Z]. Then one thinks of the theta function associated to E as a function Θ : G H [i.e., a function defined on G with values in H]. From this point of view, the various types of rigidity considered in the theory of [Mzk18] may be understood in the following fashion: (a) Cyclotomic rigidity corresponds to the portion of the tautological iso- morphism H G Gal(G/E) involving the maximal compact subgroups, def i.e., the copies of S 1 = {z C × | |z| = 1} C × . (b) Discrete rigidity corresponds to the portion of the tautological isomor- phism H G⊗Gal(G/E) involving the quotients by the maximal compact def subgroups, i.e., the copies of R >0 = {z R | z > 0} = C × /S 1 . (c) Constant rigidity corresponds to considering the normalization of Θ given by taking the values of Θ at the points of G corresponding to ± −1 to be ±1. In particular, the “canonical copy of C × that arises from (a), (b) i.e., H is related to the “copies of C × that occur as the “A p of Corollary 2.7, (e), in the following way: A p is given by the linear holomorphic automorphisms of the tangent space to a point of H. That is to say, roughly speaking, A p ( = C × ) is related to H ( = C × ) by the operation of “taking the logarithm”, followed by the operation of “taking Aut(−)” [of the resulting linearization]. Remark 2.7.3. It is interesting to note that just as the absolute Galois group G k of an MLF k may be regarded as a two-dimensional object with one rigid and one non-rigid dimension [cf. Remark 1.9.4], the topological group C × is also a two-dimensional object with one rigid dimension i.e., S 1 = {z C × | |z| = 1} C × def [a topological group whose automorphism group is of order 2] and one non-rigid dimension i.e., def R >0 = {z R | z > 0} C × [a topological group that is isomorphic to R, hence has automorphism group given by R × i.e., a “continuous family of dilations”]. Moreover, just as, in the context of Theorem 1.9, Corollary 1.10, considering G k equipped with its outer action on Δ X has the effect of rendering both dimensions of G k rigid [cf. Remark 1.9.4], considering “C × as arising, in the fashion discussed in Corollary 2.7, from a certain Aut-holomorphic orbispace has the effect of rigidifying both dimensions of C × . We refer to Remark 2.7.4 below for more on this analogy between the (i) outer action of G k on Δ X 62 SHINICHI MOCHIZUKI and the notion of an (ii) Aut-holomorphic orbispace associated to a hyperbolic orbicurve. Finally, we observe that from the point of view of the problem of finding an algorithm to construct the base field of a hyperbolic orbicurve from (i), (ii), one may think of Theorem 1.9 and Corollaries 1.10, 2.7 as furnishing solutions to various versions of this problem. Remark 2.7.4. The usual definition of a “holomorphic structure” on a Riemann surface is via local comparison to some fixed model of the topological field C. The local homeomorphisms that enable this comparison are related to one another by homeomorphisms of open neighborhoods of C that are holomorphic. On the other hand, this definition does not yield any absolute description i.e., a description that depends on mathematical structures that do not involve explicit use of models of what precisely is meant by the notion of a “holomorphic structure”. Instead, it relies on relating/comparing the given manifold to the fixed model of C an approach that is “model-explicit”. By contrast, the notion of a topological space [i.e., consisting of the datum of a collection of subsets that are to be regarded as “open”] is absolute, or “model-implicit”. In a similar vein, the approach to quasi- conformal or conformal structures via the datum of a collection of parallelograms, rectangles, or squares [cf. Propositions 2.5, 2.6; Remark 2.6.1; the theory of [Mzk14]] is “model-implicit”. The approach to “holomorphic structures” on a Riemann sur- face via the classical notion of a “conformal structure” [i.e., the datum of various orthogonal pairs of tangent vectors] is, so to speak, “relatively model-implicit”, i.e., “model-implicit” modulo the fact that it depends on the “model-explicit” definition of the notion of a differential manifold which may be thought of as a sort of “lo- cal linear structure” that is given by local comparison to the local linear structure of Euclidean space. From this point of view: The notions of an “outer action of G k on Δ X and an “Aut-holomorphic orbispace” [cf. Remark 2.7.3, (i), (ii)] have the virtue of being “model- implicit” i.e., they do not depend on any sort of [local] comparison to some fixed reference model. In this context, it is interesting to note that all of the examples given so far of “model-implicit” definitions depend on data consisting either of subsets [e.g., open subsets of a topological space; parallelograms, rectangles, or squares on a Riemann surface] or endomorphisms [e.g., the automorphisms that appear in a Galois cat- egory; the automorphisms that appear in an Aut-holomorphic structure]. [Here, in passing, we note that the appearance of “endomorphisms” in the present discussion is reminiscent of the discussion of “hidden endomorphisms” in the Introduction to [Mzk21].] Also, we observe that this dichotomy between model-explicit and model- implicit definitions is strongly reminiscent of the distinction between bi-anabelian and mono-anabelian geometry discussed in Remark 1.9.8. TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 63 Finally, we relate the archimedean theory of the present §2 to the Galois- theoretic theory of §1, in the case of number fields, via a sort of archimedean analogue of Corollary 1.10. Corollary 2.8. (Galois-theoretic Reconstruction of Aut-holomorphic Spaces) Let X, k k k NF , and 1 Δ X Π X G k 1 be as in Theorem 1.9; suppose further that k is a number field [so k NF = k], and [for simplic- ity cf. Remark 2.8.2 below] that X is a curve. Then one may think of each × archimedean prime of the field k NF {0} ( = k NF ) constructed in Theorem 1.9, × (e), as a topology on k NF {0} satisfying certain properties. Moreover, for each such archimedean prime v, there exists a functorial “group-theoretic” al- gorithm for reconstructing the Aut-holomorphic space X v associated to def X v = X × k k v × [where we write k v for the completion of k NF {0} at v] from the topological group Π X ; this algorithm consists of the following steps: (a) Define a Cauchy sequence {x j } j∈N of NF-points [of X v ] to be a se- quence of NF-points x j [i.e., conjugacy classes of decomposition groups of NF-points in Π X cf. Theorem 1.9, (a)] such that there exists a finite set of NF-points S which we shall refer to as a conductor for the Cauchy sequence satisfying the following two conditions: (i) x j ∈ S for all but finitely many j N; (ii) for every NF-rational function f on X k as in Theorem 1.9, (d), whose divisor of poles avoids S, the sequence of [non-infinite, for all but finitely many j cf. (i)] values {f (x j ) k v } j∈N forms a Cauchy sequence [in the usual sense] of k v . Two Cauchy se- quences {x j } j∈N , {y j } j∈N of NF-points which admit a common conductor S will be called equivalent if for every NF-rational function f on X k as in Theorem 1.9, (d), whose divisor of poles avoids S, the sequences of [non-infinite, for all but finitely many j] values {f (x j )} j∈N , {f (y j )} j∈N form Cauchy sequences in k v that converge to the same element of k v . For U k v an open subset and f an NF-rational function on X k as in Theorem 1.9, (d), we obtain a set N (U, f ) of Cauchy sequences of NF- points by considering the Cauchy sequences of NF-points {x j } j∈N such that f (x j ) [is finite and] U , for all j N. Then one constructs the topological space X top = X v (k v ) as the set of equivalence classes of Cauchy sequences of NF-points, equipped with the topology defined by the sets “N (U, f )”. (b) Let U X X top , U v k v be connected open subsets and f a NF- rational function on X k as in Theorem 1.9, (d), such that the func- tion defined by f on U X [i.e., by taking limits of Cauchy sequences of values in k v cf. (a)] determines a homeomorphism f U : U X U v . Write Aut hol (U v ) for the group of self-homeomorphisms U v U v (⊆ k v ), which, relative to the topological field structure of k v , can locally [on 64 SHINICHI MOCHIZUKI U v ] be expressed as a convergent power series with coefficients in k v ; A X (U X ) = f U −1 Aut hol (U v ) f U Aut(U X ). Then one constructs the Aut-holomorphic structure A X on X top as the unique [cf. Corollary 2.3, (ii)] Aut-holomorphic structure that extends the pre-Aut-holomorphic structure determined by the groups “A X (U X )”; we take X v to be the Aut- holomorphic space determined by the objects (X top , A X ). def Finally, the asserted “functoriality” is with respect to arbitrary open injective homomorphisms of profinite groups [i.e., of “Π X ”] that are compatible with the respective choices of archimedean valuations [i.e., “v”]. Proof. The validity of the algorithm asserted in Corollary 2.8 is immediate from the constructions that appear in the statement of this algorithm [together with the references quoted in these constructions]. Remark 2.8.1. One verifies immediately that the isomorphism class of the pair (1 Δ X Π X G k 1, v) depends only on the restriction of v to the subfield × k × {0} k NF {0}. Remark 2.8.2. One verifies immediately that Corollary 2.8 [as well as Corollary 2.9 below] may be extended to the case where X is a hyperbolic orbicurve that is not necessarily a curve [so X v will be an Aut-holomorphic orbispace]. Remark 2.8.3. One verifies immediately that any elliptically admissible hyper- bolic orbicurve defined over a number field is of strictly Belyi type. In particular, if one is given an elliptically admissible hyperbolic orbicurve X that is defined over a number field k, then it makes sense to apply Corollary 2.7 to the Aut-holomorphic [orbi]spaces constructed in Corollary 2.8. This compatibility between Corollaries 2.7, 2.8 [cf. also Corollary 2.9 below] is one reason why it is of interest to construct the local additive structures as in Corollary 2.7, (c), directly from the Aut-holomorphic structure as opposed to via the “parallelogram-theoretic” approach of Proposition 2.5, 2.6 [cf. also Remark 2.6.1], which is more suited to “strictly archimedean situations” i.e., situations in which one is not concerned with regarding Aut- holomorphic orbispaces as arising from hyperbolic orbicurves over number fields. Corollary 2.9. (Global-Archimedean Elliptically Admissible Compat- ibility) In the notation of Corollary 2.8, suppose further that X is elliptically admissible; take the Aut-holomorphic space X of Corollary 2.7 to be the Aut- holomorphic space determined by the objects (X top , A X ) constructed in Corollary 2.8. Then one may construct, in a functorially algorithmic fashion, an iso- morphism between the topological field k v of Corollary 2.8 and the topological fields “A p {0}” of Corollary 2.7, (e), in the following way: (a) Let x X v (k v ) be an NF-point. The local additive structures on E top [cf. Corollary 2.7, (c)] determine local additive structures on X top ; let v be an element of a sufficiently small neighborhood U X X top of x in X top TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 65 that admits such a local additive structure. Then for each NF-rational function f that vanishes at x, the assignment (v , f ) → lim n · f ( n→∞ 1 · x v ) k v n [where “· x is the operation arising from the local additive structure at x] depends only on the image df | x ω x of f in the Zariski cotangent space ω x to X v at x and, moreover, determines a topological embedding ι U X ,x : U X → Hom k v x , k v ) that is compatible with the “local additive structures” of the domain and codomain. (b) By letting the neighborhoods U X of a fixed NF-point x vary, the resulting ι U X ,x determine an isomorphism of topological fields A x  {0} k v via the condition of compatibility [with respect to the ι U X ,x ] with the nat- ural actions of A x , k v , respectively, on the domain and codomain of ι U X ,x . Moreover, as x varies, these isomorphisms are compatible with the iso- morphisms A x 1 {0} A x 2 {0} [where x 1 , x 2 X(k v ) are NF-points] of Corollary 2.7, (e). Finally, the asserted “functoriality” is to be understood in the sense described in Corollary 2.8. Proof. The validity of the algorithm asserted in Corollary 2.9 is immediate from the constructions that appear in the statement of this algorithm [together with the references quoted in these constructions]. 66 SHINICHI MOCHIZUKI Section 3: Nonarchimedean Log-Frobenius Compatibility In the present §3, we give an interpretation of the nonarchimedean local portion of the theory of §1 in terms of a certain compatibility with the “log-Frobenius func- tor” [in essence, a version of the usual “logarithm” at the various nonarchimedean primes of a number field]. In order to express this compatibility, certain abstract category-theoretic ideas which center around the notions of observables, telecores, and cores are introduced [cf. Definition 3.5]. These notions allow one to express the log-Frobenius compatibility of the mono-anabelian construction algorithms of §1 [cf. Corollary 3.6], as well as the failure of log-Frobenius compatibility that occurs if one attempts to take a “bi-anabelian” approach to the situation [cf. Corollary 3.7]. Definition 3.1. def (i) Let k be an MLF, k an algebraic closure of k, G k = Gal(k/k). Write O k k for the ring of integers of k, O k × O k for the group of units of O k , and O k  O k for the multiplicative monoid of nonzero elements [cf. [Mzk17], Example 1.1, (i)]; we shall use similar notation for other subfields of k. Let Π k be a topological group, equipped with a continuous surjection  k : Π k  G k . Note that the [p-adic, if k is of residue characteristic p] logarithm determines a Π k -equivariant isomorphism log k : k = (O × ) pf k def k [where “pf” denotes the perfection [cf., e.g., [Mzk16], §0]; the Π k -action is the action obtained by composing with  k ] of the topological group k onto the additive topological group k. Next, let us refer to an abelian monoid [e.g., an abelian group] whose subgroup of torsion elements is [abstractly] isomorphic to Q/Z as torsion- cyclotomic; let T be one of the following categories [cf. §0 for more on the prefix “ind-”]: · TF: ind-topological fields and homomorphisms of ind-topological fields; · TCG: torsion-cyclotomic ind-compact abelian topological groups and ho- momorphisms of ind-topological groups; · TLG: torsion-cyclotomic ind-locally compact abelian topological groups and homomorphisms of ind-topological groups; · TM: torsion-cyclotomic ind-topological abelian monoids and homomor- phisms of ind-topological monoids; · TS: ind-locally compact topological spaces and morphisms of ind-topological spaces; · TS: ind-locally compact abelian topological groups and homomorphisms of ind-topological groups [so we have a natural full embedding TLG → TS]. If T is equal to TF (respectively, TCG; TLG; TM; TS; TS), then let M k Ob(T) be the object determined by k (respectively, the object determined by O × ; k TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 67 × the object determined by k ; the object determined by O  ; any object of TS k equipped with a faithful continuous G k -action; any object of TS equipped with a faithful continuous G k -action). We shall refer to as a model MLF-Galois T-pair any collection of data (a), (b), (c) of the following form: (a) the topological group Π k , (b) the object M k Ob(T), (c) the action of Π k on M k [so the quotient Π k  G k may be recovered as the image of the homomorphism Π k Aut(M k ) arising from the action of (c)]; we shall often use the abbreviated notation k  M k ) for this collection of data (a), (b), (c). (ii) We shall refer to any collection of data  M ) consisting of a topological group Π, an object M Ob(T), and a continuous action of Π on M as an MLF- Galois T-pair if, for some model MLF-Galois T-pair k  M k ) [where the notation is as in (i)], there exist an isomorphism of topological groups Π k Π and an isomorphism of objects M k M of T that are compatible with the respective actions of Π k , Π on M k , M ; in this situation, we shall refer to Π as the Galois group, to the surjection Π  G determined by the action of Π on M [cf. (i)] as the Galois augmentation, to G as the arithmetic Galois group, and to M as the arithmetic data of the MLF-Galois T-pair  M ); if, in this situation, the surjection Π k  G k arises from the étale fundamental group of an arbitrary hyperbolic orbicurve (respectively, a hyperbolic orbicurve of strictly Belyi type) over k, then we shall refer to the MLF-Galois T-pair  M ) as being of hyperbolic orbicurve type (respectively, of strictly Belyi type); if, in this situation, the surjection Π k  G k is an isomorphism, then we shall refer to the MLF-Galois T-pair  M ) as being of mono-analytic type [cf. Remark 5.6.1 below for more on this terminology]. A morphism of MLF-Galois T-pairs φ : 1  M 1 ) 2  M 2 ) consists of a morphism of objects φ M : M 1 M 2 of T, together with a compatible [relative to the respective actions of Π 1 , Π 2 on M 1 , M 2 ] continuous homomorphism of topological groups φ Π : Π 1 Π 2 that induces an open injective homomor- phism between the respective arithmetic Galois groups; if, in this situation, φ M (respectively, φ Π ) is an isomorphism, then we shall refer to φ as a T-isomorphism (respectively, Galois-isomorphism). (iii) Write C T MLF for the category whose objects are the MLF-Galois T-pairs and whose morphisms are the morphisms of MLF-Galois T-pairs. Also, we shall use the same notation, except with “C” replaced by C (respectively, C; C) to denote the various subcategories determined by the T-isomorphisms (respec- tively, Galois-isomorphisms; isomorphisms); we shall use the same notation, with “MLF” replaced by MLF-hyp (respectively, MLF-sB; MLF) 68 SHINICHI MOCHIZUKI to denote the various full subcategories determined by the objects of hyperbolic orbicurve type (respectively, of strictly Belyi type; of mono-analytic type). Since × [in the notation of (i)] the formation of O  (respectively, k ; O × ; O × ) from k k × k k (respectively, O  ; O  ; k ) is clearly intrinsically defined [i.e., depends only on the k k “input data of an object of T”], we thus obtain natural functors MLF MLF C TM ; C TF MLF MLF C TM C TLG ; MLF MLF C TM C TCG ; MLF MLF C TLG C TCG i.e., by taking the multiplicative group of nonzero integral elements [i.e., the n +∞] of the elements a k × such that a −n fails to converge to 0, as N gp arithmetic data, the associated groupification M of the arithmetic data M , the subgroup of invertible elements M × of the arithmetic data M , or the maximal compact subgroups of the subgroups of the arithmetic data obtained as subgroups of invariants for various open subgroups of the Galois group. Finally, we shall write TG for the category of topological groups and continuous homomorphisms and TG TG hyp TG sB for the subcategories determined, respectively, by the étale fundamental groups of arbitrary hyperbolic orbicurves over MLF’s and the étale fundamental groups of hy- perbolic orbicurves of strictly Belyi type over MLF’s, and the homomorphisms that induce open injections on the quotients constituted by the absolute Galois groups of the base field MLF’s; also, we shall use the same notation, except with “TG” re- placed by TG to denote the various subcategories determined by the isomorphisms. Thus, for T {TF, TCG, TLG, TM, TS, TS}, the assignment  M ) → Π determines various compatible natural functors C T MLF TG [as well as double underlined versions of these functors]. (iv) Observe that [in the notation of (i)] the field structure of k determines, via the inverse morphism to log k , a structure of topological field on the topological group k . Since the various operations applied here to construct this field struc- ture on k [such as, for instance, the power series used to define log k ] are clearly intrinsically defined [cf. the natural functors defined in (iii)], we thus obtain that the construction that assigns (the ind-topological field k, with its natural Π k -action) → (the ind-topological field k , with its natural Π k -action) determines a natural functor MLF MLF log TF,TF : C TF C TF which we shall refer to as the log-Frobenius functor [cf. Remark 3.6.2 below]. Since log k determines a functorial isomorphism between the fields k, k , it follows TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 69 immediately that the functor log TF,TF is isomorphic to the identity functor [hence, in particular, is an equivalence of categories]. By composing log TF,TF with the various natural functors defined in (iii), we also obtain, for T {TLG, TCG, TM}, a functor MLF log TF,T : C TF C T MLF which [by abuse of terminology] we shall also refer to as “the log-Frobenius functor”. In a similar vein, the assignments (the ind-topological field k, with its natural Π k -action) × → (the ind-topological space k , with its natural Π k -action) (the ind-topological field k, with its natural Π k -action) × → (the ind-topological space (k ) pf , with its natural Π k -action) determine natural functors MLF MLF C TS ; λ × : C TF MLF MLF λ ×pf : C TF C TS together with diagrams of functors MLF C TF ×pf  λ MLF C TS log TF,TF −→  MLF C TF ×  λ MLF C TF ι × × λ    λ ×pf = MLF C TS MLF C TS ι log where we write ι log : λ × ◦log TF,TF λ ×pf for the natural transformation induced × by the natural inclusion “(k ) × → k = (O × ) pf → (k ) pf and ι × : λ × λ ×pf k × × for the natural transformation induced by the natural map “k (k ) pf ”. Finally, we note that the subfield of Galois-invariants “(k ) Π k of the field “k obtained by the above construction [i.e., the arithmetic data of an object in the image of the log-Frobenius functor log TF,TF ] is equipped with a natural “compactum” i.e., the compact submodule of k = (O × ) pf determined by the image of the subgroup k O k × = (O × ) Π k O × of Galois-invariants of O × which we shall refer to as the k k k pre-log-shell λ (ΠM ) log arith TF,TF ((Π  M )) MLF )] of the arithmetic data log arith [where  M ) Ob(C TF TF,TF ((Π  M )) of the object determined by applying the log-Frobenius functor log TF,TF to the object κ  M ). (v) In the notation of (i), suppose further that T {TLG, TCG, TM}; let  M ) be an MLF-Galois T-pair. Then we shall refer to the profinite Π-module def (M ) = Hom(Q/Z, M ) μ  Z  as the cyclotome associated to  M ). Also, we shall [which is isomorphic to Z] def (M ) Q/Z. write μ Q/Z (M ) = μ  Z 70 SHINICHI MOCHIZUKI (vi) Recall the “image via the Kummer map of the multiplicative group of an algebraic closure of the base field” × 1 k → lim (Π)) −→ H (J, μ  Z J [where “J” ranges over the open subgroups of Π] which was constructed via a purely “group-theoretic” algorithm in Corollary 1.10, (d), (h), for Π Ob(TG sB ). Write Anab for the category whose objects are pairs   × 1 Π, Π  {k → lim H (J, μ (Π))} −→  Z J consisting of an object Π Ob(TG sB ), together with the image of the Kummer map reviewed above, equipped with its topological field structure and natural action via Π all of which is to be understood as constructed via the “group-theoretic” algo- rithms of Corollary 1.10, (d), (h) [cf. Remark 3.1.2 below] and whose morphisms are the morphisms induced by isomorphisms of TG sB . Thus, we obtain a natural functor κ An TG sB −→ Anab which [as is easily verified] is an equivalence of categories, a quasi-inverse for which is given by the natural projection functor Anab TG sB . Remark 3.1.1. Observe that [in the notation of Definition 3.1, (i)] the topology on the field k, the groups k × and O k × , or the monoid O k  is completely determined by the field, group, or monoid structures of these objects. Indeed, the topology on O k × is precisely the profinite topology; the topologies on k, k × , and O k  are determined by the topology on the subset O k × O k  k × k [cf. the various natural functors of Definition 3.1, (iii); the fact that O k × k × may be characterized as the subgroup of elements divisible by arbitrary powers of some prime number]. Suppose that T  = TS, TS. Then note that one may apply this observation to the various subfields, subgroups, or submonoids obtained from the arithmetic data of an MLF-Galois T-pair by taking the invariants with respect to some open subgroup of the Galois group. Thus, we conclude that one obtains an entirely equivalent theory if one omits the specification of the topology, as well as of the “ind-” structure [i.e., one works with the inductive limit fields, groups, or monoids, as opposed to the inductive systems of such objects] from the objects of T considered in Definition MLF is precisely the data 3.1, (i). In particular, the data that forms an object of C TM used to construct the “model p-adic Frobenioids” of [Mzk17], Example 1.1. Remark 3.1.2. It is important to note that, by definition, the algorithms of Corollary 1.10 form an essential portion of each object of the category Anab. Put another way, the “software” constituted by these algorithms is not just executed once, leaving behind some “output data” that suffices for the remainder of the development of the theory, but rather executed over and over again within each object of Anab. TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 71 Remark 3.1.3. One natural variant of the notion of an “MLF-Galois T-pair of hyperbolic orbicurve type” is the notion of an “MLF-Galois T-pair of tempered hyperbolic orbicurve type”, i.e., the case where [in the notation of Definition 3.1, (ii)] Π k  G k arises from the tempered fundamental group of a hyperbolic orbicurve over k [cf. Remarks 1.9.1, 1.10.2]. We leave to the reader the routine details of developing the resulting tempered version of the theory to follow. Proposition 3.2. (Monoid Cyclotomes and Kummer Maps) Let T {TM, TF};  M T ) Ob(C MLF ) an MLF-Galois T-pair, with arithmetic Ga- T lois group Π  G. Write  M TM ) Ob(C MLF TM ) for the object obtained from  M T ) by applying the identity functor if T = TM or by applying the natural functor of Definition 3.1, (iii), if T = TF. Then: (i) The arguments given in the proof of [Mzk9], Proposition 1.2.1, (vii), yield a functorial [i.e., relative to C MLF , in the evident sense cf. Remark 3.2.2 below] T algorithm for constructing the natural isomorphism  (M TM )) Z H 2 (G, μ  Z i.e., by composing the natural isomorphism [of “Brauer groups”] gp ) H 2 (G, μ Q/Z (M TM )) H 2 (G, M TM [where “gp” denotes the groupification of a monoid] with the inverse of the natural isomorphism [of “Brauer groups”] gp unr gp ) ) H 2 (G, M TM ) H 2 (G unr , (M TM unr [where M TM M TM denotes the submonoid of elements fixed by the kernel of the quotient G  G unr of Corollary 1.10, (b)] followed by the natural composite isomorphism unr gp unr gp unr ×  Z) H 2 (G unr , (M TM ) ) H 2 (G unr , (M TM ) /(M TM ) ) H 2 ( Z, Q/Z [where “×” denotes the subgroup of invertible elements of a monoid; the isomor- unr gp unr × ) /(M TM ) Z is obtained by considering a generator of the phism (M TM  unr unr × monoid M TM /(M TM ) = N; we apply the isomorphism G unr Z of Corollary 1.10, (b)] and then applying the functor Hom(Q/Z, −) to the resulting isomorphism H 2 (G, μ Q/Z (M TM )) Q/Z [cf. also Remark 3.2.1 below]. (ii) By considering the action of open subgroups H Π on elements of M TM H that are roots of elements of M TM [i.e., the submonoid of M TM consisting of H- invariant elements], we obtain a functorial [i.e., relative to C MLF , in the evident T sense] algorithm for constructing the Kummer maps H M TM H 1 (H, μ  (M TM )); Z 1 M TM lim (M TM )) −→ H (J, μ  Z J where “J” ranges over the open subgroups of Π. In particular, the “μ  (M TM )” in Z (G)” [cf. Remarks 3.2.1, 3.2.2 below]; if, the above display may be replaced by “μ  Z 72 SHINICHI MOCHIZUKI moreover,  M T ) is of hyperbolic orbicurve type, then the “μ  (M TM )” in Z (Π)” [cf. Corollary 1.10, (c)] or “μ κ (Π)” the above display may be replaced by “μ  Z  Z [cf. Remark 1.10.3, (ii)] cf. Remark 3.2.2 below. (iii) Suppose that  M T ) is of strictly Belyi type. Then the construction of Corollary 1.10, (h), determines an additive structure [hence, in particular, a topological field structure] on the union with “{0}” of the group generated by the image of the Kummer map 1 (M TM )) M TM lim −→ H (J, μ  Z J , of (ii). In particular, these constructions yield a functorial [i.e., relative to C MLF T in the evident sense cf. Remark 3.2.2 below] algorithm for constructing this topological field structure. ), then the natural functor of Definition 3.1, (iv) If  M T ) Ob(C MLF T (iii), induces an injection Isom C MLF ((Π  M T ),  M T )) → Isom TG (Π, Π ) T on sets of isomorphisms; this injection is a bijection if T = TM, or if [T is either TM or TF, and]  M T ),  M T ) are of strictly Belyi type. In particular, if  M T ) is of hyperbolic orbicurve type, then the group Aut C MLF ((Π  M T )) T which is isomorphic to a subgroup of Aut TG (Π) that contains the subgroup of Aut TG (Π) determined by the inner automorphisms of Π is center-free; the categories TG hyp , TG sB , TG hyp , TG sB , C MLF-hyp , C MLF-sB , C MLF-hyp , C MLF-sB are T T T T id-rigid [cf. §0]. (v) The algorithm of (iii) yields a natural [1-]factorization MLF-sB C TF −→ C T MLF-sB log T,T  −→ C T MLF-sB  where T  {TF, TLG, TCG, TM}; the first arrow is the natural functor of Def- inition 3.1, (iii), if T = TM, or the identity functor if T = TF of the [“sB” MLF C T MLF of Definition versions of the] log-Frobenius functors log TF,T  : C TF  3.1, (iv). Moreover, the functor log T,T is isomorphic to the identity functor [hence, in particular, is an equivalence of categories]. Proof. Assertions (i), (ii), (iii), (v) are immediate from the constructions that appear in the statement of these assertions [together with the references quoted in these constructions]. The injectivity portion of assertion (iv) follows from the functorial algorithms of assertions (i), (ii) [which imply that automorphisms of  M TM ) that act trivially on Π necessarily act trivially on M TM ]. In light of this injectivity, the center-free-ness portion of assertion (iv) follows immediately from the slimness of Π [cf., e.g., [Mzk20], Proposition 2.3, (ii)]. The surjectivity portion of assertion (iv) follows from assertion (iii), when  M T ),  M T ) TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 73 are of strictly Belyi type, and from considering the “copy of M TM O  embedded k in abelianizations of open subgroups of G G k via local class field theory” [cf., e.g., [Mzk9], Proposition 1.2.1, (iii), (iv)], together with assertions (i), (ii) [cf. also the first displayed isomorphism of Corollary 1.10, (b)], when T = TM. Remark 3.2.1. Note that the algorithm applied to construct the natural isomor- phism of Corollary 1.10, (a), is essentially the same as the algorithm of Proposition 3.2, (i). In particular, this algorithm does not require that  M T ) be of hy- perbolic orbicurve type. Thus, by imposing the condition of “compatibility with the natural isomorphism of Corollary 1.10, (a)”, we thus obtain, in the context of Proposition 3.2, (i), a functorial algorithm for constructing the natural isomorphism μ  (M TM ) μ  (G) Z Z [cf. also Remark 1.10.3, (ii)]. Remark 3.2.2. Note that [cf. Remark 1.10.1, (iii)] the functoriality of Propo-  sition 3.2, (i), when applied to the isomorphism H 2 (G, μ  (M TM )) Z, is to be Z  that ap- understood in the sense of a “compatibility”, relative to dividing the Z” pears as the codomain of these isomorphisms by a factor given by the index of the image of the induced open homomorphism on arithmetic Galois groups [cf. Defini- tion 3.1, (ii)]. A similar remark [cf. Remark 1.10.1, (i)] applies to the cyclotome (Π)” that appears in Proposition 3.2, (ii). We leave the routine details to the “μ  Z reader. In a similar vein, one may consider Kummer maps for “O × [as opposed to “O  ”], in which case the natural isomorphism of Remark 3.2.1 is only determined  × -multiple [cf. [Mzk17], Remark 2.4.2]. up to a Z Proposition 3.3. (Unit Kummer Maps) Let T {TLG, TCG}. Let  MLF M ) Ob(C T ) be an MLF-Galois T-pair, with arithmetic Galois group Π  G. Then: (i) By considering the action of open subgroups H Π on elements of M that are roots of elements of M H [i.e., the subgroup of M consisting of H-invariant elements], we obtain a functorial [i.e., relative to C MLF , in the evident sense] T algorithm for constructing the Kummer maps (M )); M H H 1 (H, μ  Z 1 M lim (M )) −→ H (J, μ  Z J where “J” ranges over the open subgroups of Π. In this situation [unlike the situation of Proposition 3.2, (ii)], the natural isomorphism μ  (M ) μ  (G) [cf. Z Z ×  -)multiple if Remark 3.2.1] is only determined up to a {±1}- (respectively, Z T = TLG (respectively, T = TCG) [cf. (ii) below; [Mzk17], Remark 2.4.2, in the case T = TCG]. 74 SHINICHI MOCHIZUKI (ii) If  M ) Ob(C MLF ), then any isomorphism  M )  T M ) induces isomorphisms Π Π , μ  (M ) μ  (M ), which determine an in- Z Z jection ((Π  M ),  M )) → Isom TG (Π, Π ) × Isom TG  (M ), μ  (M )) Isom C MLF Z Z T which is a bijection if T = TCG. If T = TLG, then the homomorphism ((Π  M ),  M )) Isom TG (Π, Π ) is surjective, with fibers of Isom C MLF T cardinality two. Proof. The portion of assertion (i) concerning Kummer maps is immediate from the definitions and the references quoted. The portion of assertion (i) concerning (M ) μ  (G) follows by observing that the algorithm of Propo- the isomorphism μ  Z Z sition 3.2, (i) [cf. also Remark 3.2.1] may be applied, up to a {±1}- (respectively,  × -)indeterminacy, if T = TLG (respectively, T = TCG). The injectivity portion Z of assertion (ii) follows from assertion (i) via a similar argument to the argument used to derive the injectivity portion of Proposition 3.2, (iv), from Proposition 3.2, (i), (ii); the surjectivity onto Isom TG (Π, Π ) follows from a similar argument to the argument applied to prove the surjectivity portion of Proposition 3.2, (iv). If T = TCG (respectively, T = TLG), then the remainder of assertion (ii) follows by  × on M (respectively, observing that observing that there is a natural action of Z × as soon as an automorphism of  M ) preserves the submonoid O  k = M k × [i.e., preserves the “positive elements” of k /O × = Q], one may apply the functorial algorithm of Proposition 3.2, (i)). k Lemma 3.4. (Topological Distinguishability of Additive and Mul- tiplicative Structures) In the notation of Definition 3.1, (i), let α : k × k × be an automorphism of the topological group k × , α pf : (k × ) pf (k × ) pf the automorphism induced on the perfection. Then α pf ((O k  ) pf ) ⊆ (O k × ) pf . Proof. Indeed, since O k × is easily verified to be the maximal compact subgroup of k × , α induces an isomorphism O k × O k × . Thus, α pf ((O k × ) pf ) = (O k × ) pf , so an in- clusion α pf ((O k  ) pf ) (O k × ) pf would imply that (O k  ) pf (O k × ) pf , a contradiction. Definition 3.5. (i) We shall refer to as a diagram of categories D = (  Γ D , {D v }, {D e }) any collection of data as follows: (a) an oriented graph  Γ D [cf. §0]; (b) for each vertex v of  Γ D , a category D v ; (c) for each edge e of  Γ D that runs from a vertex v 1 to a vertex v 2 , a functor D e : D v 1 D v 2 . Let D = (  Γ D , {D v }, {D e }) be a diagram of categories. Then observe that any path [γ] [cf. §0] on  Γ D that runs from a vertex v 1 to a vertex v 2 determines i.e., TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 75 by composing the various functors “D e ”, for edges e that appear in this path a functor D [γ] : D v 1 D v 2 . We shall refer to the diagram of categories E obtained by restricting the data of D to an oriented subgraph  Γ E of  Γ D as a subdiagram of categories of D. (ii) Let D = (  Γ D , {D v }, {D e }) be a diagram of categories. Then we shall refer to as a family of homotopies H = (E H , }) on D any collection of data as follows: (a) a saturated [cf. §0] set E H Ω(  Γ D ) × Ω(  Γ D ) of ordered pairs of paths on  Γ D , which we shall refer to as the boundary set of the family of homotopies H; we shall refer to every path on  Γ D that occurs as a component of an element of E H as a boundary set path; (b) for each  = ([γ 1 ], 2 ]) E H , a natural transformation ζ : D 1 ] D 2 ] which we shall refer to as a homotopy from 1 ] to 2 ] such that the following conditions are satisfied: ζ ([γ],[γ]) is the identity natural transformation for each ([γ], [γ]) E H ; if  = ([γ 1 ], 2 ]),   = ([γ 2 ], 3 ]), and   = ([γ 1 ], 3 ]) belong to E H , then ζ  = ζ  ζ ; if, for some 3 ], 4 ] Ω(  Γ D ), the pairs  = ([γ 1 ], 2 ]) and   = ([γ 3 ]◦[γ 1 ]◦[γ 4 ], 3 ]◦ 2 ] 4 ]) belong to E H , then ζ  = D 3 ] ζ D 4 ] . If, in this situation, E H is the smallest [cf. §0] saturated subset of Ω(  Γ D ) × Ω(  Γ D ) E H , then we shall say that the family of ho- that contains a given subset E H . We motopies H = (E H , }) is generated by the homotopies indexed by E H shall refer to a family of homotopies H = (E H , }) on D as symmetric if E H is symmetrically saturated [cf. §0]. [Thus, if H = (E H , }) is symmetric, then every ζ is an isomorphism.] We shall refer to a collection of families of homo- topies {H ι = (E H ι , ι ι })} ι∈I on D as being compatible if there exists a family of homotopies H = (E H , }) on D such that, for each ι I,  ι E H ι , we have E H ι E H and ζ ι ι = ζ ι . (iii) Let D = (  Γ D , {D v }, {D e }) be a diagram of categories. Then we shall refer to as an observable S = (S, v S , H) [on D] any collection of data as follows: (a) a diagram of categories S = (  Γ S , {S v }, {S e }) that contains D as a subdi- agram of categories [so  Γ D  Γ S ]; (b) a vertex v S of  Γ S , which we shall refer to as the observation vertex, such that the set of vertices of  Γ S \  Γ D is equal to {v S }, and, moreover, every edge of  Γ S \  Γ D runs from a vertex of  Γ D to v S ; (c) a family of homotopies H on S such that every boundary set path of H has terminal vertex equal to v S . Let S = (S, v S , H) be an observable on D. Then we shall say that S is symmetric if H is symmetric. We shall say that S is a core [on D] if the boundary set of H is equal to the set of all co-verticial pairs of paths on the underlying oriented graph  Γ S of S with terminal vertex equal to v S [which implies that S is symmetric], and, moreover, every vertex of  Γ D appears as the initial vertex of a path on  Γ S with terminal vertex equal to v S . Suppose that S is a core on D. Then we shall refer to 76 SHINICHI MOCHIZUKI the observation vertex of S as the core vertex of S and [by abuse of terminology, when there is no fear of confusion] to S v S as a “core on D”. (iv) Let S = (S, v S , H) be a core on a diagram of categories D = (  Γ D , {D v }, {D e }). Then we shall refer to as a telecore T = (T , J ) on D over the core S any collection of data as follows: (a) a diagram of categories T = (  Γ T , {T v }, {T e }) that contains S as a subdi- agram of categories [so  Γ D  Γ S  Γ T ] such that  Γ T ,  Γ S have the same vertices, and, moreover, every edge of  Γ T \  Γ S runs from v S to a vertex of  Γ D ; we shall refer to such edges of  Γ T as the telecore edges; (b) J is a family of homotopies on T such that J | S = H whose boundary set is equal to the subset of Ω(  Γ T ) of pairs ([γ 3 ] 1 ], 3 ] 2 ]), where ([γ 1 ], 2 ]) is a co-verticial pair of paths on  Γ T with terminal vertex equal to v S , and 3 ] is a path on  Γ T with initial vertex equal to v S . In this situation, a family of homotopies H cnct on T that is compatible with J will be referred to as a contact structure for the telecore T . (v) Let D = (  Γ D , {D v }, {D e }) and D  = (  Γ D  , {D  v  }, {D e  }) be diagrams of categories. Then a 1-morphism of diagrams of categories Φ : D D  is defined to be a collection of data follows: (a) a morphism of oriented graphs Φ Γ :  Γ D  Γ D  ; (b) for each vertex v of  Γ D , a functor Φ v : D v D  Φ  Γ (v) ; (c) for each edge e of  Γ D that runs from a vertex v 1 to a vertex v 2 , an isomorphism of functors Φ e : D  Φ  Γ (e) Φ v 1 Φ v 2 D e . A 2-morphism Θ : Φ Ψ between 1-morphisms Φ, Ψ : D D  such that Φ Γ = Ψ Γ is defined to be a collection of natural transformations v : Φ v Ψ v }, where v ranges over the vertices of  Γ D , such that Ψ e (D  Φ  Γ (e) Θ v 1 ) = v 2 D e ) Φ e : D  Φ  Γ (e) Φ v 1 Ψ v 2 D e for each edge e of  Γ D that runs from a vertex v 1 to a vertex v 2 . We shall say that a 1-morphism Φ : D D  is an equivalence of diagrams of categories if there exists a 1-morphism Ψ : D  D such that Ψ Φ, Φ Ψ are [2-]isomorphic to the respective identity 1-morphisms of D, D  . If D (respectively, D  ) is equipped with a family of homotopies H (respectively, H  ), then we shall say that an equivalence Φ : D D  is compatible with H, H  if Φ Γ induces a bijection between the boundary sets of H, H  , and, moreover, the natural transformations that constitute H, H  [cf. the data of (ii), (b)] are compatible [in the evident sense] with the natural transformations that constitute Φ [cf. the data (c) in the above definition]; in this situation, one verifies immediately that if Φ is compatible with H, H  , then so is any equivalence Ψ : D D  that is isomorphic to Φ. We shall say that D is vertex- rigid (respectively, edge-rigid) if, for every vertex v (respectively, edge e) of  Γ D , the TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 77 category D v (respectively, the functor D e ) is id-rigid (respectively, rigid) [cf. §0]. If D is vertex-rigid and edge-rigid, then we shall say that D is totally rigid. Thus, if D is edge-rigid, then any equivalence Φ : D D  is completely determined by Φ Γ and the v } [i.e., the data (a), (b) in the above definition]. In a similar vein, if D is vertex-rigid, then any two isomorphic equivalences Φ, Ψ : D D  admit a unique [2-]isomorphism Φ Ψ. In particular, if D is vertex-rigid, then it is natural to speak of the automorphism group Aut(D) of D, i.e., the group determined by the isomorphism classes of self-equivalences of D. (vi) Let D = (  Γ D , {D v }, {D e }) be a diagram of categories;  a nexus of  Γ D [cf. §0]; D ≤ , D ≥ the subdiagrams of categories determined, respectively, by the pre- and post-nexus portions of  Γ D [cf. §0]. Then we shall say that D is totally -rigid if the pre-nexus portion D ≤ is totally rigid. Let us suppose that D is totally -rigid. Write Aut  (D ≤ ) Aut(D ≤ ) for the subgroup of isomorphism classes of self-equivalences of D ≤ that preserve  and induce a self-equivalence of D  that is isomorphic to the identity self- equivalence. Let Φ ≤ : D ≤ D ≤ be a self-equivalence whose isomorphism class ≤ ] Aut  (D ≤ ). Then Φ ≤ extends naturally to an equivalence Φ : D D which is the identity on  Γ D ≥ and which associates to each vertex v  =  of  Γ D ≥ the identity self-equivalence of D v . [Here, we observe that the isomorphism of func- tors of (v), (c), is naturally determined by the isomorphism of ≤ )  with the identity self-equivalence of D  .] Moreover, this assignment Φ ≤ → Φ clearly maps isomorphic equivalences to isomorphic equivalences and is compatible with composition of equivalences. In particular, this assignment yields a natu- ral “action” of the group Aut  (D ≤ ) on D. We shall refer to the resulting self- equivalences of D as nexus self-equivalences of D [relative to the nexus ] and the resulting classes of self-equivalences of D [i.e., arising from isomorphism classes of “Φ ≤ ”] as nexus-classes of self-equivalences of D [relative to the nexus ]. Remark 3.5.1. If one just works with diagrams of categories without considering any observables, then it is difficult to understand the “global structure” of the diagram since [by definition!] it does not make sense to speak of the relationship between objects that belong to different categories [e.g., at distinct vertices of the diagram]. Thus: The notion of an observable may be thought of as a sort of “partial pro- jection of the dynamics of a diagram of categories” onto a single category, within which it makes sense to compare objects that arise from distinct categories at distinct vertices of the diagram. Moreover: A core on a diagram of categories may be thought of as an extraction of a certain portion of the data of the objects at the various categories in the 78 SHINICHI MOCHIZUKI diagram that is invariant with respect to the “dynamics” arising from the application of the various functors in the diagram. Put another way, one may think of a core as a sort of “constant portion” of the diagram that lies, in a consistent fashion, “under the entire diagram” [cf. the use of the term “core” in the theory of [Mzk11], §2]. Then: A telecore may be thought of as a sort of partial section i.e., given by the telecore edges of the “structure morphisms to the core” which does not disturb the coricity [i.e., the property of being a core] of the original core. Moreover, although, in the definition of a telecore, we do not assume the existence of families of homotopies that guarantee the compatibility of applying composites of functors by traveling along arbitrary co-verticial pairs of paths emanating from the core vertex, any failure of such a compatibility may always be eliminated in a fashion reminiscent of a “telescoping sum” by projecting back down to the core vertex [cf. the discussion of Remark 3.6.1, (ii), (c), below]. Put another way, one may think of a telecore as a device that satisfies a sort of “time lag compatibility”, i.e., as a device whose “compatibility apparatus” does not go into operation immediately, but only after a certain “time lag” [arising from the necessity to travel back down to the core vertex]. Also, for more on the meaning of cores and telecores, we refer to Remark 3.6.5 below. The terminology of Definition 3.5 makes it possible to formulate the first main result of the present §3. Corollary 3.6. (MLF-Galois-theoretic Mono-anabelian Log-Frobenius Compatibility) Write def ; X = C MLF-sB T def E = TG sB ; def MLF-sB N = C TS where [in the notation of Definition 3.1] T {TM, TF}. Consider the diagram of categories D log log −→ X −→ X ... ... X id  id −1 ... ...   id +1  X λ ×   λ ×pf N  E κ  An Anab  E TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 79 where we use the notation “log”, “λ × ”, “λ ×pf for the evident [double-underlined/ overlined] restrictions of the arrows “log T,T ”, “λ × ”, “λ ×pf of Definition 3.1, (iv) [cf. also Proposition 3.2, (iii), (v)]; for positive integers n 6, we shall denote by D ≤n the subdiagram of categories of D determined by the first n [of the six] rows of D; we write L for the countably ordered set determined [cf. §0] by the infinite linear oriented graph  Γ opp D ≤1 [so the elements of L correspond to vertices of the first row of D] and L = L {} def for the ordered set obtained by appending to L a formal symbol  [which we think of as corresponding to the unique vertex of the second row of D] such that  < , for all  L; id  denotes the identity functor at the vertex  L; the notation “. . . denotes an infinite repetition of the evident pattern. Then: (i) For n = 4, 5, 6, D ≤n admits a natural structure of core on D ≤n−1 . That is to say, loosely speaking, E, Anab “form cores” of the functors in D. (ii) The assignments   Π, Π  {k → lim (Π))} →  O k  ), −→ H (J, μ  Z × 1  k ×  {0}) J determine [i.e., for each choice of T] a natural “forgetful” functor Anab φ An −→ X which is an equivalence of categories, a quasi-inverse for which is given by the composite π An : X Anab of the natural projection functor X E with κ An : E Anab; write η An : φ An π An id X for the isomorphism arising from the “group-theoretic” algorithms of Corollary 1.10 [cf. also Proposition 3.2, (ii), (iii)]. Moreover, φ An gives rise to a telecore structure T An on D ≤4 , whose underlying diagram of categories we denote by D An , by appending to D ≤5 telecore edges ... φ +1 ... X Anab φ    log −→ Anab X φ  −→  φ −1 ... X ... log −→ X from the core Anab to the various copies of X in D ≤2 given by copies of φ An , which we denote by φ  , for  L . That is to say, loosely speaking, φ An determines a 0 ] for the path telecore structure on D ≤4 . Finally, for each  L , let us write  1 on  Γ D An of length 0 at  and  ] for the path on  Γ D An of length {4, 5} [i.e., depending on whether or not  = ] that starts from , descends [say, via λ × ] to the core vertex “Anab”, and returns to  via the telecore edge φ  . Then the collection of natural transformations −1 −1  , η  , η  , η  } ∈L,∈L 80 SHINICHI MOCHIZUKI where we write η  for the identity natural transformation from the arrow φ  : Anab X to the composite arrow id  φ  : Anab X and η  : (D An )  1 ] (D An )  0 ] for the isomorphism arising from η An generate a contact structure H An on the telecore T An . (iii) The natural transformations ι log, : λ × id  log λ ×pf id +1 , ι × : λ × λ ×pf [cf. Definition 3.1, (iv)] belong to a family of homotopies on D ≤3 that determines on D ≤3 a structure of observable S log on D ≤2 and, moreover, is compatible with the families of homotopies that constitute the core and telecore structures of (i), (ii). (iv) The diagram of categories D ≤2 does not admit a structure of core on D ≤1 which [i.e., whose constituent family of homotopies] is compatible with [the constituent family of homotopies of ] the observable S log of (iii). Moreover, the telecore structure T An of (ii), the contact structure H An of (ii), and the observable S log of (iii) are not simultaneously compatible [but cf. Remark 3.7.3, (ii), below]. (v) The unique vertex  of the second row of D is a nexus of  Γ D . More- over, D is totally -rigid, and the natural action of Z on the infinite linear oriented graph  Γ D ≤1 extends to an action of Z on D by nexus-classes of self- equivalences of D. Finally, the self-equivalences in these nexus-classes are com- patible with the families of homotopies that constitute the cores and observ- able of (i), (iii); these self-equivalences also extend naturally [cf. the technique of extension applied in Definition 3.5, (vi)] to the diagram of categories [cf. Definition 3.5, (iv), (a)] that constitutes the telecore of (ii), in a fashion that is compatible with both the family of homotopies that constitutes this telecore structure [cf. Definition 3.5, (iv), (b)] and the contact structure H An of (ii). Proof. In the following, if φ is a functor appearing in D, then let us write [φ] for the path on the underlying oriented graph  Γ D of D determined by the edge corresponding to φ [cf. §0]. Now assertion (i) is immediate from the definitions and the fact that the algorithms of Corollary 1.10 are “group-theoretic” in the sense that they are expressed in language that depends only on the profinite group given as “input data”. Next, we consider assertion (ii). The portion of assertion (ii) concerning φ An and T An is immediate from the definitions and the “group-theoretic” algorithms of Corollary 1.10 [cf. also Proposition 3.2, (ii), (iii)]. Thus, it suffices to show the existence of a contact structure H An as described. To this end, let us first ob- serve that the isomorphism of log with the identity functor [cf. Definition 3.1, (iv); Proposition 3.2, (v)] is compatible [in the evident sense] with the natural tranforma- −1 −1 , η  , η  } ∈L,∈L . On the other hand, this compatibility implies tions  , η  that one may, in effect, “contract” D ≤2 down to a single vertex [equipped with the TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 81 category X ] and the various paths from  to Anab down to a single edge i.e., that, up to redundancies, one is, in effect, dealing with a diagram of categories with two vertices “X and “Anab” joined by two [oriented] edges φ An , π An . Now the existence of a family of homotopies that contains the collection of natural transfor- −1 } ∈L,∈L follows immediately. This completes the proof of mations  , η  , η  assertion (ii). Next, we consider assertion (iii). Write E log for the set of ordered pairs of paths on  Γ D ≤3 [i.e., the underlying oriented graph of D ≤3 ] consisting of pairs of paths of the following three types: (1) ([λ × ] [id  ] [log] [γ], ×pf ] [id +1 ] [γ]), where [γ] is a path on D ≤3 whose terminal vertex lies in the first row of D ≤3 ; (2) ([λ × ] [γ], ×pf ] [γ]), where [γ] is a path on D ≤3 whose terminal vertex lies in the second row of D ≤3 ; (3) ([γ], [γ]), where [γ] is a path on D ≤3 whose terminal vertex lies in the third row of D ≤3 . Then one verifies immediately that E log satisfies the conditions (a), (b), (c), (d), (e) given in §0 for a saturated set. Moreover, the natural transformation(s) ι log, (respectively, ι × ) determine(s) the homotopies for pairs of paths of type (1) (re- spectively, (2)). Thus, we obtain an observable S log , as desired. Moreover, it is immediate from the definitions i.e., in essence, because the various Galois groups that appear remain “undisturbed” by the various manipulations involving arithmetic data that arise from “ι log, ”, “ι × that this family of homotopies is compatible with the families of homotopies that constitute the core and telecore structures of (i), (ii). This completes the proof of assertion (iii). Next, we consider assertion (iv). Suppose that D ≤2 admits a structure of core on D ≤1 in a fashion that is compatible with the observable S log of (iii). Then this core structure determines, for  L, a homotopy ζ 0 for the pair of paths ([id +1 ], [id  ] [log]); thus, by composing the result ζ 0  of applying λ × to ζ 0 with the homotopy ζ 1 associated [via S log ] to the pair of paths ([λ × ]◦[id  ]◦[log], ×pf ]◦ [id +1 ]) [of type (1)], we obtain a natural transformation ζ 1  = ζ 1 ζ 0  : λ × id +1 λ ×pf id +1 which, in order for the desired compatibility to hold, must coincide with the homotopy ζ 2 associated [via S log ] to the pair of paths ([λ × ]◦[id +1 ], ×pf ]◦[id +1 ]) [of type (2)]. On the other hand, by writing out explicitly the meaning of such an equality ζ 1  = ζ 2 , we conclude that we obtain a contradiction to Lemma 3.4. This completes the proof of the first incompatibility of assertion (iv). The proof of the second incompatibility of assertion (iv) is entirely similar. That is to say, if we compose on the right with +1 ] the various paths that appeared in the proof of the first incompatibility, then in order to apply the argument applied in the proof of the first incompatibility, it suffices to relate the paths [id +1 ] +1 ]; [id  ] [log] +1 ] 82 SHINICHI MOCHIZUKI [a task that was achieved in the proof of the first incompatibility by applying the core structure whose existence was assumed in the proof of the first incompatibility]. In the present situation, applying the homotopy η +1 of the contact structure H An yields a homotopy  ]  [id +1 ]◦[φ +1 ]; on the other hand, we obtain a homotopy 1  ]  [id  ]  ]  [id  ]  ] [log] +1 ]  [id  ] [log] +1 ] by applying the homotopy η  of the contact structure H An , followed by the homo- topies of the telecore T An , followed by the homotopy η  of the contact structure H An . Thus, by applying the argument applied in the proof of the first incompatibility, we obtain two mutually contradictory homotopies × ]◦[φ  ]  ×pf ]◦[id +1 ]◦[φ +1 ]. This completes the proof of the second incompatibility of assertion (iv). Finally, we consider assertion (v). The total -rigidity in question follows immediately from Proposition 3.2, (iv) [cf. also the final portion of Proposition 3.2, (v)]. The remainder of assertion (v) follows immediately from the definitions. This completes the proof of assertion (v). Remark 3.6.1. (i) The “output” of the “log-Frobenius observable” S log of Corollary 3.6, (iii), may be summarized intuitively in the following diagram: ... Π +1  ... Π +1  × k +1 × (k +1 ) pf Π  Π   ←  × k  ... × (k  ) pf Π −1 Π −1  ... ← × k −1 ...  × (k −1 ) pf ... × where the arrows “→” are the natural morphisms [cf. ι × !]; k  , for  L, is a × copy of “k that arises, via id  , from the vertex  of D ≤1 ; the arrows “←” are the × inclusions arising from the fact that k  is obtained by applying the log-Frobenius × functor log to k +1 [cf. ι log, !]; the isomorphic “Π  ’s” that act on the various × k  ’s and their perfections correspond to the coricity of E [cf. Corollary 3.6, (i)]. Finally, the incompatibility assertions of Corollary 3.6, (iv), may be thought of as a statement of the non-existence of some “universal reference model” × k model × that maps isomorphically to the various k  ’s in a fashion that is compatible with the various arrows “→”, “←” of the above diagram cf. also Corollary 3.7, (iv), below. TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 83 (ii) In words, the content of Corollary 3.6 may be summarized as follows [cf. the “intuitive diagram” of (i)]: (a) The Galois groups that act on the various objects under consideration are compatible with all of the operations involved in particular, the op- erations constituted by the functors log, κ An , φ An and the various related families of homotopies cf. the coricity asserted in Corollary 3.6, (i). (b) By contrast, the operation constituted by the log-Frobenius functor [as “observed” via the observable S log ] is not compatible with the field struc- ture of the fields [i.e., “k”] involved [cf. Corollary 3.6, (iv)]. (c) As a consequence of (a), the “group-theoretic reconstruction” of the base field via Corollary 1.10 is compatible with all of the operations involved, except “momentarily” when log acts on the output of φ An an operation which “temporarily obliterates” the field structure of this output, although this field structure may be recovered by projecting back down to E [cf. (a)] and applying κ An . This sort of “conditional compatibility” i.e., up to a “brief temporary exception” is expressed in the telecoricity asserted in Corollary 3.6, (ii). In particular, if one thinks of the various operations involved as being “software” [cf. Remark 1.9.8], then the projection to E i.e., the operation of looking at the Galois group is compatible with simultaneous execution of all the “software” [in particular, including log!] under consideration; the “group-theoreticity” of the algorithms of Corollary 1.10 implies that Anab, κ An satisfy a similar “compatibility with simultaneous execution of all software” [cf. Remark 3.1.2] property. Remark 3.6.2. (i) The reasoning that lies behind the name “log-Frobenius functor” may be understood as follows. At a very naive level, the natural logarithm may be thought of as a sort of “raising to the -th power” [where  0 is some indefinite positive infinitesimal] i.e., “” plays the role in characteristic [ →]0 of “p” in charac- teristic p > 0. More generally, the logarithm frequently appears in the context of Frobenius actions, in particular in discussions involving canonical coordinates, such as in [Mzk1], Chapter III, §1. (ii) In general, Frobenius morphisms may be thought of as “compression mor- phisms”. For instance, this phenomenon may be seen in the most basic example of a Frobenius morphism in characteristic p > 0, i.e., the morphism t → t p on F p [t]. Put another way, the “compression” operation inherent in a Frobenius morphism may be thought of as an approximation of some sort of “absolute con- stant object” [such as F p ]. In the context of the log-Frobenius functor, this sort of compression phenomenon may be seen in the pre-log-shells defined in Definition 3.1, (iv), which will play a key role in the theory of §5 below. 84 SHINICHI MOCHIZUKI (iii) Whereas the log-Frobenius functor obliterates the field [or ring] structure [cf. Remark 3.6.1, (ii), (b)] of the fields involved, the usual Frobenius morphism in positive characteristic is compatible with the ring structure of the rings involved. On the other hand, unlike generically smooth morphisms, the Frobenius morphism in positive characteristic has the effect of “obliterating the differentials” of the schemes involved. Remark 3.6.3. The diagram D of Corollary 3.6 cf., especially, the first two rows D ≤2 and the various natural actions of Z discussed in Corollary 3.6, (v) may be thought of as a sort of combinatorial version of G m cf. the point of view of Remark 1.9.7. Remark 3.6.4. One verifies immediately that one may give a tempered version of Propositions 3.2, 3.3; Corollary 3.6 [cf. Remarks 1.9.1, 1.10.2]. Remark 3.6.5. The notions of “core” and “telecore” are reminiscent of certain aspects of “Hensel’s lemma” [cf., e.g., [Mzk21], Lemma 2.1]. That is to say, if one compares the successive approximation operation applied in the proof of Hensel’s lemma [cf., e.g., the proof of [Mzk21], Lemma 2.1] to the various operations [in the form of functors] that appear in a diagram of categories, then one is led to the following analogy: cores ←→ sets of solutions of “étale”, i.e., “slope zero” equations telecores ←→ sets of solutions of “positive slope” equations i.e., where one thinks of applications of Hensel’s lemma in the context of mixed characteristic, so the property of being “étale in characteristic p > 0” may be regarded as corresponding to “slope zero”. That is to say, the “étale case” of Hensel’s lemma is the easiest to understand. In this “étale case”, the invertibil- ity of the Jacobian matrix involved implies that when one executes each successive approximation operation, the set of solutions lifts uniquely, i.e., “transports isomor- phically” through the operation. This sort of “isomorphic transport” is reminiscent of the definition of a core on a diagram of categories. On the other hand, the “pos- itive slope case” of Hensel’s lemma is a bit more complicated [cf., e.g., the proof of [Mzk21], Lemma 2.1]. That is to say, although the set of solutions does not quite “transport isomorphically” in the simplest most transparent fashion, the fact that the Jacobian matrix involved is invertible up to a factor of p implies that the set of solutions “essentially transports isomorphically, up to a brief temporary lag” cf. the “brief temporary exception” of Remark 3.6.1, (ii), (c). Put another way, if one thinks in terms of connections on bases on which p is nilpotent, in which case formal integration takes the place of the “successive approximation operation” of Hensel’s lemma, then one has the following analogy: cores ←→ vanishing p n -curvature telecores ←→ nilpotent p n -curvature [where we refer to [Mzk4], Chapter II, §2; [Mzk7], §2.4, for more on “p n -curvature”] cf. Remark 3.7.2 below. TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 85 Remark 3.6.6. (i) In the context of the analogy between telecores and “positive slope situa- tions” discussed in Remark 3.6.5, one question that may occur to some readers is the following: What are the values of the positive slopes implicitly involved in a telecore? At the time of writing, it appears to the author that, relative to this analogy, one should regard telecores as containing “all positive slopes”, or, alternatively, “positive slopes of an indeterminate nature”, which one may think of as corresponding to the lengths of the various paths emanating from the core vertex that one may travel along before descending back down to the core [cf. Remark 3.5.1]. Indeed, from the point of view of the analogy [discussed in Remark 3.7.2 below] with the uniformizing MF -objects constructed in [Mzk1], [Mzk4], this is natural, since uniformizing MF -objects also involve, in effect, “all positive slopes”. Moreover, since telecores are of an “abstract, combinatorial nature” [cf. Remark 1.9.7] i.e., not of a “linear, module-theoretic nature”, as is the case with MF -objects it seems somewhat natural that this “combinatorial non-linearity” should interfere with any attempts to “separate out the various distinct positive slopes”, via, for instance, a “linear filtration”, as is often possible in the case of MF -objects. (ii) From the point of view of the analogy with [uniformizing] MF -objects, one has the following [rough] correspondence: slope zero ←→ Frobenius “” (an isomorphism) positive slope ←→ Frobenius “” p n · (an isomorphism) [where “” is to be understood as shorthand for the phrase “acts via”; n is a positive integer]. Perhaps the most fundamental example in the p-adic theory of such a [uniformizing] MF -object arises from the p-adic Galois representation obtained by extracting p-power roots of the standard unit U on the multiplicative group G m over Z p , in which case the “positive slope” involved corresponds to the action dlog(U ) = dU/U → p · dlog(U ) induced by the Frobenius morphism U → U p . In the situation of Corollary 3.6, an analogue of this sort of correspondence may be seen in the “temporary failure of coricity” [cf. Remark 3.6.1, (ii), (c); the failure of coricity documented in Corollary 3.6, (iv)] of the “mono-anabelian telecore” of Corollary 3.6, (ii). That is to say, mutiplication by a positive power of p corresponds precisely to this “temporary failure of coricity”, a failure that is remedied [where the “remedy” corresponds to the isomorphism that appears by “pealing off an appropriate power of p] by projecting back down to Anab, an operation which [in light of the “group-theoretic nature” of the algorithms applied in κ An ] induces an isomorphism of, for instance, the base-field reconstructed after the application of log with the base-field that was reconstructed prior to the application of log. Remark 3.6.7. Note that in the situation of Corollary 3.6 [cf. also the ter- minology introduced in Definition 3.5], although we have formulated things in the 86 SHINICHI MOCHIZUKI language of categories and functors, in fact, the mathematical constructs in which we are ultimately interested have much more elaborate structures than categories and functors. That is to say: In fact, what we are really interested in is not so much “categories” and “functors”, but rather “types of data” and “operations” [i.e., algorithms!] that convert some “input type of data” into some “output type of data”. One aspect of this state of affairs may be seen in the fact that the crucial functors log, κ An of Corollary 3.6 are equivalences of categories [which, moreover, are, in certain cases, isomorphic to the identity functor! cf. Definition 3.1, (iv), (vi)] i.e., from the point of view of the purely category-theoretic structure [cf., e.g., the point of view of [Mzk14], [Mzk16], [Mzk17]!] of “X ”, “E”, or “Anab”, these functors are “not doing anything”. On the other hand, from the point of view of “types of data” and “operations” on these “types of data” [cf. Remark 3.6.1], the operations constituted by the functors log, κ An are highly nontrivial. To some extent, this state of affairs may be remedied by working with appropriate observables [i.e., which serve to project the operations constituted by functors between different categories down into arrows in a single category cf. Remark 3.5.1], as in Corollary 3.6, (iii), (iv). Nevertheless, the use of observables does not constitute a fundamental solution to the issue raised above. It is the hope of the author to remedy this state of affairs in a more definitive fashion in a future paper by introducing appropriate “enhancements” to the usual theory of categories and functors. To understand what is gained in Corollary 3.6 by the mono-anabelian theory of §1, it is useful to consider the following “bi-anabelian analogue” of Corollary 3.6. Corollary 3.7. (MLF-Galois-theoretic Bi-anabelian Log-Frobenius In- compatibility) In the notation and conventions of Corollary 3.6, suppose, further, that T = TF. Consider the diagram of categories D ... ... X × E X pr +1  log X −→ X × E X pr   log X −→ X × E X ...  pr −1 ... X λ ×   λ ×pf N  E where the second to fourth rows of D are identical to the second to fourth rows of the diagram D of Corollary 3.6; D ≤1 is obtained by applying the “categorical fiber product” (−) × E X [cf. §0] to D ≤1 ; pr  denotes the projection to the first factor on the copy of X × E X at the vertex  L. Also, let us write π  : X × E X X TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 87 for the projection to the second factor on the copy of X × E X at the vertex  L, D for the result of appending these arrows π  to D where we think of the codomain “X of the arrows π  as a new “core” vertex lying in the first row of D “under” the various copies of “X × E X at the vertices of L and D ≤n [where n {1, 2, 3, 4}] for the subdiagram of D constituted by D ≤n , together with the newly appended arrows π  . Then: (i) D = D ≤4 (respectively, D ≤1 ) admits a natural structure of core on D ≤3 (respectively, D ≤1 ). That is to say, loosely speaking, E “forms a core” of the func- tors in D ; the “second factor” X “forms a core” of the functors in D ≤1 . [Thus, we think of the second factor of the various fiber product categories X × E X as being a “universal reference model” cf. Remark 3.7.3 below.] (ii) Write δ X : X X × E X for the natural diagonal functor and bi θ bi : pr bi 1 pr 2 bi for the isomorphism between the two projection functors pr bi 1 , pr 2 : X × E X X that arises from the functoriality i.e., the bi-anabelian [or “Grothendieck Conjecture”-type] portion [cf. Remark 1.9.8] of the “group-theoretic” algorithms of Corollary 1.10. Then δ X is an equivalence of categories, a quasi-inverse for which is given by the projection to the second factor π X : X × E X X ; θ bi determines an isomorphism θ X : δ X π X id X × E X . Moreover, δ X gives rise to a telecore structure T δ on D ≤1 , whose underlying diagram of categories we denote by D δ , by appending to D ≤1 telecore edges X δ    ... δ +1 ... X × E X log X −→ X × E X  δ −1 log X −→ X × E X ... ... given by copies of δ X , from the core X to the various copies of X × E X in D ≤1 which we denote by δ  , for  L. That is to say, loosely speaking, δ X determines a telecore structure on D ≤1 . Finally, let us write D for the diagram of categories obtained by gluing [in the evident sense] D δ to D along D ≤1 and then appending an edge δ  X −→ X from the core vertex of D δ to the vertex at  [i.e., the unique vertex of the second row of D ] given by a copy of the identity functor; for each  L, let us write 0 1  ] for the path on  Γ D of length 0 at  and  ] for the path on  Γ D of length 88 SHINICHI MOCHIZUKI 2 that starts from , descends via π  to the core vertex, and returns to  via the telecore edge δ  . Then the collection of natural transformations −1 −1 , θ  , θ  } ∈L  , θ  where we write θ  for the identity natural transformation from the arrow δ  : X X to the composite arrow pr  δ  : X X and θ  : D 1 ] D 0 ]   for the isomorphism arising from θ X generate a family of homotopies H δ on D [hence, in particular, by restriction, a contact structure on the telecore admits a natural structure of core on D ≤3 in a fashion T δ ]. Finally, D = D ≤4 compatible with the core structure of D ≤4 on D ≤3 discussed in (i) [that is to say, loosely speaking, E “forms a core” of the functors in D ]. (iii) Write ι log, : λ × pr  log X λ ×pf pr +1 , ι × = ι × : λ × λ ×pf def for the natural transformations determined by the natural transformations of Corol- lary 3.6, (iii). Then these natural transformations ι log, , ι × belong to a family of that determines on D ≤3 a structure of observable S log on homotopies on D ≤3 D ≤2 and, moreover, is compatible with the families of homotopies that constitute the core and telecore structures of (i), (ii). (iv) The diagram of categories D ≤2 does not admit a structure of core on D ≤1 which [i.e., whose constituent family of homotopies] is compatible with [the constituent family of homotopies of ] the observable S log of (iii). Moreover, the telecore structure T δ of (ii), the family of homotopies H δ of (ii), and the observable S log of (iii) are not simultaneously compatible. (v) The vertex  of the second row of D is a nexus of  Γ D . Moreover, D is totally -rigid, and the natural action of Z on the infinite linear oriented graph  Γ extends to an action of Z on D by nexus-classes of self-equivalences of D ≤1 D . Finally, the self-equivalences in these nexus-classes are compatible with H δ [cf. (ii)], as well as with the families of homotopies that constitute the cores, telecore, and observable of (i), (ii), (iii). Proof. The proofs of the various assertions of the present Corollary 3.7 are entirely similar to the proofs of the corresponding assertions of Corollary 3.6. Remark 3.7.1. In some sense, the purpose of Corollary 3.7 is to examine what happens if the mono-anabelian theory of §1 is not available, i.e., if one is in a situation in which one may only apply the bi-anabelian version of this theory. This is the main reason for our assumption that “T = TF” in Corollary 3.7 that is TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 89 to say, when T = TM, one is obliged to apply Proposition 3.2, (v), a result whose proof requires one to invoke the mono-anabelian theory of §1. Remark 3.7.2. The “Frobenius-theoretic” point of view of Remarks 3.6.2, 3.6.5, 3.6.6 motivates the following observation: The situation under consideration in Corollaries 3.6, 3.7 is structurally reminiscent of the situation encountered in the p-adic crystalline theory, for instance, when one considers the MF -objects of [Falt]. That is to say, the core E plays the role of the “absolute constants”, given, for instance, in the p-adic theory by [absolutely] unramified extensions of Z p . The isomorphism bi θ bi : pr bi 1 pr 2 bi [cf. Corollary 3.7, (ii)] between the two projection functors pr bi 1 , pr 2 : X × E X X is formally reminiscent of the notion of a(n) [integrable] connection in the crystalline theory. The diagonal functor δ X : X X × E X [cf. Corollary 3.7, (ii)] is formally reminiscent of the diagonal embedding into the divided power envelope of the product of a scheme with itself in the crystalline theory. Moreover, since, in the crystalline theory, this divided power envelope may itself be regarded as a crystal, the [various divided powers of the ideal defining the] diagonal embedding may then be regarded as a sort of Hodge filtration on this crys- tal. That is to say, the telecore structure of Corollary 3.7, (ii), may be regarded as corresponding to the Hodge filtration, or, for instance, in the context of the theory of indigenous bundles [cf., e.g., [Mzk1], [Mzk4]], to the Hodge section. Thus, from this point of view, the second incompatibility of assertion (iv) of Corollaries 3.6, 3.7, is reminiscent of the fact that [in general] the Frobenius action on the crystal underlying an MF -object fails to preserve the Hodge filtration. For instance, in the case of indigenous bundles, this failure to preserve the Hodge section may be regarded as a consequence of the isomorphicity of the Kodaira-Spencer morphism associated to the Hodge section. On the other hand, the log-Frobenius-compatibility of the mono-anabelian models discussed in Corollary 3.6 may be regarded as corre- sponding to canonical Frobenius actions on the crystals constituted by the divided power envelopes discussed above cf. the uniformizing MF -objects constructed in [Mzk1], [Mzk4]. Moreover, the compatibility of the coricity of E, Anab with the telecore and contact structures of Corollary 3.6, (ii), on the one hand, and the “log- Frobenius observable” S log of Corollary 3.6, (iii), on the other hand, is reminiscent of the construction of the Galois representation associated to an MF -object by considering the submodule that lies in the 0-th step of the Hodge filtration and, moreover, is fixed by the action of Frobenius [cf. Remark 3.7.3, (ii), below]. Thus, in summary, we have the following “dictionary”: the coricity of E ←→ absolutely unramified constants bi-anabelian isomorphism of projection functors ←→ integrable connections diagonal functor δ X telecore str. ←→ Hodge filtration/section 90 SHINICHI MOCHIZUKI bi-anabelian log-incompatibility ←→ Kodaira-Spencer isomorphism “forgetful” functor φ An telecore str. ←→ underlying vector bundle of MF -object mono-anabelian log-compatibility ←→ [positive slope!] uniformizing MF -objects [where “str.” stands for “structure”]. This analogy with MF -objects will be developed further in §5 below. Remark 3.7.3. The significance of Corollary 3.7 in the context of our discussion of the mono-anabelian versus the bi-anabelian approach to anabelian geometry [cf. Remark 1.9.8] may be understood as follows. (i) We begin by considering the conditions that we wish to impose on the framework in which we are to work. First of all, we wish to have some sort of fixed reference model of “X ”. The fact that this model is to be fixed throughout the discussion then translates into the requirement that this copy of X be a core, relative to the various “operations” performed during the discussion. On the other hand, one does not wish for this model to remain “completely unrelated to the operations of interest”, but rather that it may be related, or compared, to the various copies of this model that appear as one executes the operations of interest. In our situation, we wish to be able to relate the “fixed reference model” to the copies of this model i.e., “log-subject copies” that are subject to the log-Frobenius operation [i.e., functor cf. Remark 3.6.7]. Moreover, since the log-Frobenius functor is isomorphic to the identity functor [cf. Proposition 3.2, (v)], hence may only be “properly understood” in the context of the natural transformations “ι × and “ι log ”, we wish for everything that we do to be compatible with the operation of “making an observation” via these natural transformations. Thus, in summary, the main conditions that we wish to impose on the framework in which we are to work are the following: (a) coricity of the model; (b) comparability of the model to log-subject copies of the model; (c) consistent observability of the various operations executed [especially log]. In the context of the various assertions of Corollaries 3.6, 3.7, these three aspects correspond as follows: (a) ←→ the coricity of (i), the “coricity portion” of the telecore structure of (ii), (b) ←→ the telecore and contact structures/families of homotopies of (ii), (c) ←→ the “log-observable” of (iii). [Here, we refer to the content of Definition 3.5, (iv), (b), as the “coricity portion” of a telecore structure.] In the case of Corollary 3.7, the “fixed reference model” is realized by applying a “category-theoretic base-change” (−) × E X , as in Corollary 3.7, i.e., the copy of “X used to effect this base-change serves as the “fixed reference model”; in the case of Corollary 3.6, the “fixed reference model” is given by “Anab” [i.e., especially, the second piece of parenthesized data “(−, −)” in the definition of Anab cf. Definition 3.1, (vi)]. Also, we observe that the second incompatibility of TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 91 assertion (iv) of Corollaries 3.6, 3.7 asserts, in effect, that neither of the approaches of these two corollaries succeeds in simultaneously realizing conditions (a), (b), (c), in the strict sense. (ii) Let us take a closer look at the mono-anabelian approach of Corollary 3.6 from the point of view of the discussion of (i). From the point of view of “operations performed”, this approach may be summarized as follows: One starts with “Π”, applies the mono-anabelian algorithms of Corollary 1.10 to obtain an object of Anab, then forgets the “group-theoretic origins” of such objects to obtain an object of X [cf. the telecore structure of Corollary 3.6, (ii)], which is subject to the action of log; this action of log obliterates the ring structure [indeed, it obliterates both the additive and multiplicative structures!] of the arithmetic data involved, hence leaving behind, as an invariant of log, only the original “Π”, to which one may again apply the mono-anabelian algorithms of Corollary 1.10. Π    Π  Π × k An × k    Π log Π   × k An The point of the mono-anabelian approach is that although log obliterates the ring structures involved [cf. the second incompatibility of Corollary 3.6, (iv)], E i.e., “Π” remains constant [up to isomorphism] throughout the application of the various operations; this implies that the “purely group-theoretic constructions” of Corollary 1.10 i.e., Anab, κ An also remain constant throughout the application of the various operations. In particular, in the above diagram, despite the fact × that log obliterates the ring structure of “k  ”, the operations executed induce an isomorphism between all the “Π’s” that appear, hence an isomorphism between the × initial and final “(Π  k An )’s”. At a more technical level, this state of affairs may be witnessed in the fact that although [cf. the proof of the second incompatibility of Corollary 3.6, (iv)] there exist incompatible composites of homotopies involving the families of homotopies that constitute the telecore, contact, and observable structures involved, these composites become compatible as soon as one augments the various paths involved with a path back down to the core vertex “Anab”. At a more philosophical level: This state of affairs, in which the application of log does not immediately × yield an isomorphism of “k ’s”, but does after “pealing off the operation × of forgetting the group-theoretic construction of k An ”, is reminiscent of the situation discussed in Remark 3.6.6, (ii), concerning “Frobenius  p n · (an isomorphism)” [i.e., where Frobenius induces an isomorphism after “pealing off an ap- propriate power of p]. (iii) By contrast, the bi-anabelian approach of Corollary 3.7 may be understood in the context of the present discussion as follows: One starts with an arbitrarily 92 SHINICHI MOCHIZUKI × declared “model” copy “Π  k model of X , then forgets the fact that this copy was arbitrarily declared a model [cf. the telecore structure of Corollary 3.7, (ii)]; this × yields a copy “Π  k  of X on which log acts in a fashion that obliterates the ring structure of the arithmetic data involved, hence leaving behind, as an invariant of log, only the original “Π”. Π Π    × k model × k    Π log Thus, unlike the case with the mono-anabelian approach, if one tries to work with × × another model “Π  k model after applying log, then the “k model portion of this × new model cannot be related to the “k model portion of the original model in a con- sistent fashion i.e., such a relation is obstructed by log, which obliterates the ring × structure of k model . Moreover, unlike the case with the mono-anabelian approach, there is “no escape route” in the bi-anabelian approach [i.e., which requires the use of models] from this situation given by taking a path back down to some core vertex [i.e., such as “Anab”]. Relative to the analogy with usual Frobenius actions [cf. Remark 3.6.6, (ii)], this situation is reminiscent of the Frobenius action on the ideal defining the diagonal of a divided power envelope I⊆O PD S × S [where S is, say, smooth over F p ] i.e., Frobenius simply maps I to 0 in a fashion that does not allow one to “recover, in an isomorphic fashion, by pealing off a power of p”. [Indeed, the data necessary to “peal off a power of p” consists, in essence, of a Frobenius lifting which is, in essence, equivalent to the datum of a uniformizing MF -object cf. Remark 3.7.2; the theory of [Mzk1], [Mzk4].] In particular: Although it is difficult to give a completely rigorous formulation of the ? question “bi-anabelian =⇒ mono-anabelian” raised in Remark 1.9.8, the state of affairs discussed above strongly suggests a negative answer to this question. (iv) The following questions constitute a useful guide to understanding better the gap that lies between the “success of the mono-anabelian approach” and the “failure of the bi-anabelian approach”, as documented in (i), (ii), (iii): (a) In what capacity i.e., as what type of mathematical object [cf. Re- mark 3.6.7] does one transport i.e., “effect the coricity of” [cf. con- × dition (a) of (i)] the fixed reference model of “k down to “future log-generations” [i.e., smaller elements of L]? (b) On precisely what type of data [cf. Remark 3.6.7] does the comparison [cf. condition (b) of (i)] via telecore/contact structures depend? That is to say, in the mono-anabelian approach, the answer to both questions is given by E [i.e., “Π”], Anab; by contrast, in the bi-anabelian approach, the answer TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 93 × to (b) necessarily requires the inclusion of the “model k model a requirement that is incompatible with the coricity required by (a) [i.e., since log obliterates the ring × structure of k model ]. ( Galois [core] )   Galois + log-subject ( Galois [of model] ) arithmetic data Fig. 1: Mono-anabelian comparison only requires “Galois input data”. ( Galois [core] )  Galois + model arithmetic data  ? ... ... Galois + log-subject arithmetic data Fig. 2: Bi-anabelian comparison requires “arithmetic input data”. One way to understand this state of affairs is as follows. If one attempts to construct a “bi-anabelian version of Anab”, then the requirement of coricity means that the × “model k model employed in the bi-anabelian reconstruction algorithm of such a “bi-anabelian version of Anab” must be compatible with the various isomorphisms × × k model k  of Remark 3.6.1, (i) where we recall that the various distinct × k  ’s are related to one another by log i.e., compatible with the “building”, or × “edifice”, of k  ’s constituted by these isomorphisms together with the diagram of Remark 3.6.1, (i). That is to say, in order for the required coricity to hold, this bi-anabelian reconstruction algorithm must be such that it only depends on × the ring structure of k model “up to log” i.e., the algorithm must be immune to the confusion [arising from log] of the additive and multiplicative structures that constitute this ring structure. On the other hand, the bi-anabelian approach to reconstruction clearly does not satisfy this property [i.e., it requires that the ring × structure of k model be left intact]. Remark 3.7.4. In the context of the issue of distinguishing between the mono- anabelian and bi-anabelian approaches to anabelian geometry [cf. Remark 3.7.3], one question that is often posed is the following: Why can’t one somehow sneak a “fixed refence model” into a “mono- anabelian reconstruction algorithm” by finding, for instance, some copy of Q or Q p inside the Galois group “Π” and then building up some copy of the hyper- bolic orbicurve under consideration over this base field [i.e., this copy of Q, 94 SHINICHI MOCHIZUKI Q p ], which one then takes as one’s “model”, thus allowing one to “reduce” mono-anabelian problems to bi-anabelian ones [cf. Remark 1.9.8]? One important observation, relative to this question, is that although it is not so difficult to “construct” such copies of Q or Q p from Π, it is substantially more difficult to construct copies of the algebraic closures of Q or Q p in such a way that the resulting absolute Galois groups are isomorphic to the appropriate quotient of the given Galois group “Π” in a functorial fashion [cf. Remark 3.7.5 below]. Moreover, once one constructs, for instance, a universal pro-finite étale covering of an appropriate hyperbolic orbicurve on which Π acts [in a “natural”, functorial fashion], one must specify [cf. question (a) of Remark 3.7.3, (iv)] in what capacity i.e., as what type of mathematical object one transports [i.e., “effects the coricity of”] this pro-hyperbolic orbicurve model down to “future log-generations”. Then if one only takes a naive approach to these issues, one is ultimately led to the arbitrary introduction of “models” that fail to be immune to the application of the log-Frobenius functor that is to say, one finds oneself, in effect, in the situation of the “bi-anabelian approach” discussed in Remark 3.7.3. Thus, the above discussion may be summarized in flowchart form, as follows: construction of model [universal pro-covering] schemes without essential use of Π natural functorial action of Π on model scheme is trivial must supplement model scheme with Π Gal(model scheme) essentially equivalent situation to “bi-anabelian approach”. Put another way, if one tries to sneak a “fixed refence model” that may be con- structed without essential use of “Π” into a “mono-anabelian reconstruction algo- rithm”, then one finds oneself confronted with the following two mutually exclusive choices concerning the type of mathematical object [cf. question (a) of Remark 3.7.3, (iv)] that one is to assign to this model: (∗) the model arises from “Π” =⇒ “functorially trivial model”; (∗∗) the model does not arise from “Π” =⇒ “bi-anabelian approach”. In particular, Figures 1 and 2 of Remark 3.7.3, (iv), are not [at least in an “a priori sense”] “essentially equivalent”. Remark 3.7.5. (i) From the point of view of “constructing models of the base field from Π” [cf. the discussion of Remark 3.7.4], one natural approach to the issue of finding TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 95 “Galois-compatible models” is to work with Kummer classes of scheme-theoretic functions, since Kummer classes are tautologically compatible with Galois actions. [Indeed, the use of Kummer classes is one important aspect of the theory of §1.] Moreover, in addition to being “tautologically Galois-compatible”, Kummer classes also have the virtue of fitting into a container def (Π)) H(Π) = H 1 (Π, μ  Z [cf. Corollary 1.10, (d), where we take “Π X to be Π] which inherits the coricity of Π [cf. question (a) of Remark 3.7.3, (iv)] in a very natural, tautological fashion. Thus, once one characterizes, in a “group-theoretic” fashion, the “Kummer subset” of this container H(Π) [i.e., the subset constituted by the Kummer classes that arise from scheme-theoretic functions], it remains to reconstruct the additive structure on [the union with {0} of] the set of Kummer classes [cf. the theory of §1]. If one takes the point of view of the question posed in Remark 3.7.4, then it is tempting to try to use “models” solely as a means to reconstruct this additive structure. This approach, which combines the “purely group-theoretic” [i.e., “mono- anabelian”] container H(Π) with the indirect use of “models” to recon- struct the additive structure [or the Kummer subset], may be thought of as a sort of intermediate alternative between the “mono-anabelian” and “bi-anabelian” approaches discussed so far; in the discussion to follow, we shall refer to this sort of intermediate approach as “pseudo-mono- anabelian”. With regard to implementing this pseudo-mono-anabelian approach, we observe that the “automorphism version of the Grothendieck Conjecture” [i.e., the functo- riality of the algorithms of Corollary 1.10, applied to automorphisms] allows one to conclude that the additive structure “pulled back from a model scheme via the Kummer map” is rigid [i.e., remains unaffected by automorphisms of Π]. On the other hand, the “isomorphism version of the Grothendieck Conjecture” [i.e., the functoriality of the algorithms of Corollary 1.10, applied to isomorphisms cf. the isomorphism θ bi of Corollary 3.7, (ii)] allows one to conclude that this additive structure is independent of the choice of model. (ii) The pseudo-mono-anabelian approach gives rise to a theory that satisfies many of the useful properties satisfied by the mono-anabelian theory. Thus, at first glance, it is tempting to consider simply abandoning the mono-anabelian approach, in favor of the pseudo-mono-anabelian approach. Closer inspection reveals, how- ever, that the situation is not so simple. Indeed, relative to the coricity requirement of Remark 3.7.3, (i), (a), there is no problem with allowing the “hidden models” on which the pseudo-mono-anabelian approach depends in an essential way to remain hidden. On the other hand, the issue of relating [cf. Remark 3.7.3, (i), (b)] these hidden models to log-subject copies of these models is more complicated. Here, the central problem may be summarized as follows [cf. Remark 3.7.3, (iv), (a), (b)]: Problem (∗ type ): Find a type of mathematical object that [in the context of the framework discussed in Remark 3.7.3, (i)] serves as a common type of mathematical object for both “coric models” and “log-subject copies”, 96 SHINICHI MOCHIZUKI thus rendering possible the comparison of “coric models” and “log-subject copies”. That is to say, in the mono-anabelian approach, this “common type” is furnished by the objects that constitute E and [in light of the “group-theoreticity” of the algo- rithms of Corollary 1.10] Anab; in the bi-anabelian approach, the “common object” is furnished by the “copy of X that appears in the base-change (−) × E X ”. Note that it is precisely the existence of this “common type of mathematical object” that renders possible the definition of the telecore structure cf., especially, the functor δ X of Corollary 3.7, (ii). In particular, we note that the definition of the diagonal functor δ X is possible precisely because of the equality of the types of mathematical object involved in the two factors of X × E X . On the other hand, if, in imple- menting the pseudo-mono-anabelian approach, one tries to use, for instance, E [i.e., without including the “hidden model”!], then although this yields a framework in which it is possible to work with the “mono-anabelian container H(Π)”, this does not allow one to describe the contents [i.e., the Kummer subset with its ring struc- ture] of this container. That is to say, if one describes these “contents” via “hidden models”, then the data contained in the “common type” is not sufficient for the operation of relating this description to the “conventional description of contents” that one wishes to apply to the log-subject copies. Indeed, if the coric models and log-subject copies only share the container H(Π), but not the description of its contents i.e., the description for the coric models is some “mysterious descrip- tion involving hidden models”, while the description for the log-subject copies is the “standard Kummer map description” then, a priori, there is no reason that these two descriptions should coincide. For instance, if the “mysterious description” is not related to the “standard description” via some common description applied to a common type of mathematical object, then, a priori, the “mysterious description” could be [among a vast variety of possibilities] one of the following: (1) Instead of embedding the [nonzero elements of the] base field into H(Π) via the usual Kummer map, one could consider the embedding obtained by composing the usual Kummer map with the automorphism induced by some automorphism of the quotient Π  G k [cf. the notation of Corollary 1.10, where we take “Π X to be Π] which is not of scheme-theoretic origin [cf., e.g., [NSW], the Closing Remark preceding Theorem 12.2.7]. (2) Alternatively, instead of embedding the function field of the curve under consideration into H(Π) via the usual Kummer map, one could consider the embedding obtained by composing the usual Kummer map with the  × . automorphism of H(Π) given by muliplication by some element Z Thus, in order to ensure that such pathologies do not arise, it appears that there is little choice but to include the ring/scheme-theoretic models in the common type that one adopts as a “solution to (∗ type )”, so that one may apply the “standard Kummer map description” in a simultaneous, consistent fashion to both the coric data and the log-subject data. But [since these models are “functorially obstructed from being subsumed into Π” cf. Remark 3.7.4] the inclusion of such ring/scheme- theoretic models amounts precisely to the “bi-anabelian approach” discussed in Remark 3.7.3 [cf., especially, Figure 2 of Remark 3.7.3, (iv)]. TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 97 (iii) From a “physical” point of view, it is natural to think of data that sat- isfies some sort of coricity such as the étale fundamental group Π as being × “massless”, like light. By comparison, the arithmetic data “k on which the log-Frobenius functor acts non-isomorphically may be thought of as being like matter which has “weight”. This dichotomy is reminiscent of the dichotomy dis- cussed in the Introduction to [Mzk16] between “étale-like” and “Frobenius-like” structures. Thus, in summary: coricity, “étale-like” structures ←→ massless, like light “Frobenius-like” structures ←→ matter of positive mass. From this point of view, the discussion of (i), (ii) may be summarized as follows: Even if the container H(Π) is massless, it one tries to use it to carry “cargo of substantial weight”, then the resulting package [of container plus cargo] is no longer massless. On the other hand, the very existence of mono-anabelian algorithms as discussed in §1, §2 corresponds, in this analogy, to the “conversion of light into matter” [cf. the point of view of Remark 1.9.7]! (iv) Relative to the dichotomy discussed in the Introduction to [Mzk16] between “étale-like” and “Frobenius-like” structures, the problem observed in the present paper with the bi-anabelian approach may be thought of as an example of the phenomenon of the non-applicability of Galois [i.e., “étale-like”] descent with respect to “Frobenius-like” morphisms [i.e., the existence of descent data for a “Frobenius- like” morphism which cannot be descended to an object on the codomain of the morphism]. In classical arithmetic geometry, this phenomenon may be seen, for instance, in the non-descendability of Galois-invariant coherent ideals with respect to morphisms such as Spec(k[t]) Spec(k[t n ]) [where n 2 is an integer; k is a field], or [cf. the discussion of “X × E X in Remark 3.7.2] the difference between an integrable connection and an integrable connection equipped with a compatible Frobenius action [e.g., of the sort that arises from an MF -object]. Remark 3.7.6. With regard to the pseudo-mono-anabelian approach discussed in Remark 3.7.5, one may make the following further observations. (i) In order to carry out the pseudo-mono-anabelian approach [or, a fortiori, the mono-anabelian approach], it is necessary to use the full profinite étale fundamental group of a hyperbolic orbicurve, say, of strictly Belyi type. That is to say, if, for instance, one attempts to use the geometrically pro-Σ fundamental group of a hyperbolic curve [i.e., where Σ is a set of primes which is not equal to the set of all primes], then the crucial injectivity of the Kummer map [cf. Proposition 1.6, (i)] fails to hold. In particular, this failure of injectivity means that one cannot work with the crucial additive structure on [the union with {0} of] the image of the Kummer map. (ii) In a similar vein, if one attempts to work, for instance, with the absolute Galois group of a number field i.e., in the absence of any geometric fundamental group of a hyperbolic orbicurve over the number field then, in order to work with Kummer classes, one must contend with the nontrivial issue of finding an  appropriate [profinite] cyclotome [i.e., copy of Z(1)”] to replace the “curve-based cyclotome M X of Proposition 1.4, (ii) [cf. also Remark 1.9.5]. 98 SHINICHI MOCHIZUKI (iii) Next, we observe that if one attempts to construct “models of the base field” via the theory of “characters of qLT-type” as in [Mzk20], §3 [cf. also the theory of [Mzk2], §4], then although [just as was the case with Kummer classes] such “qLT-models of the base field” are tautologically Galois-compatible and admit a coricity inherited from the coricity of Π, [unlike the case with Kummer classes] the essential use of p-adic Hodge theory implies that the resulting “construction of the base field” [cf. the discussion of Remark 1.9.5] is incompatible with the operation of passing from global [e.g., number] fields to local fields [i.e., does not admit an analogue of the first portion of Corollary 1.10, (h)], hence also incompatible with the operation of relating the resulting “constructions of the base field”at different localizations of a number field. Such localization [i.e., in the terminology of §5, “panalocalization”] properties will play a key role in the theory of §5 below. (iv) In the context of (iii), it is interesting to note that geometrically pro-Σ fundamental groups as in (i) also fail to be compatible with localization. Indeed, even if some sort of pro-Σ analogue of the theory of §1 is, in the future, obtained for the primes lying over prime numbers Σ, such an analogue is impossible at the primes lying over prime numbers ∈ Σ [since, as is easily verified, at such primes, the automorphisms of G k [notation of Corollary 1.10] that are not of scheme-theoretic origin may extend, in general, to automorphisms of the full arithmetic [geometri- cally pro-Σ] fundamental group]. (v) At the time of writing, it appears to be rather difficult to give a mono- anabelian “group-theoretic” algorithm as in Theorem 1.9 in the case of num- ber fields by somehow “gluing together” [mono-anabelian, “group-theoretic”] al- gorithms [cf. the approach via p-adic Hodge theory discussed in (iii)] applied at nonarchimedean completions of the number field. That is to say, if one tries, for instance, to construct a number field F as a subset of the product of copies of F constructed at various nonarchimedean completions of F , then it appears to be a highly nontrivial issue to reconstruct the correspondences between the various “local copies” of F . Indeed, if one attempts to work with abelianizations of local and global Galois groups and apply class field theory [i.e., in the fashion of [Uchi], in the case of function fields], then one may only recover the “global copy” of F × embedded in the idèles up to an indeterminacy that involves, in particular, various “solenoids” [cf., e.g., [ArTt], Chapter Nine, Theorem 3]. On the other hand, if one attempts to work with local and global Kummer classes, then one must contend with the phenomenon that it is not clear how to lift local Kummer classes to global Kummer classes; that is to say, the indeterminacies that occur for such liftings are of a nature roughly reminiscent of the global Kummer classes whose vanishing is implied by the so-called Leopoldt Conjecture [i.e., in its formulation concerning p-adic localizations of units of a number field], which is unknown in general at the time of writing. Remark 3.7.7. (i) One way to interpret the fact that the log-Frobenius operation log is not a ring homomorphism [cf. the discussion of Remarks 3.7.3, 3.7.4, 3.7.5] is to think of TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 99 “log” as constituting a sort of “wall” that separates the two “distinct scheme theories” that occur before and after its application. The étale fundamental groups that arise in these “distinct scheme theories” thus necessarily correspond to dis- out tinct, unrelated basepoints. Thus, if, for i = 1, 2, G i  Δ i is a copy of the out outer Galois action on the geometric fundamental group “G k  Δ X of Theo- rem 1.9 that arises in one of these two “distinct scheme theories” separated by the “log-wall”, then although this log-wall cannot be penetrated by ring structures [i.e., by “scheme theory”], it can be penetrated by the abstract profinite group structure of the G i cf. the Galois-equivariance of the map “log k of Definition 3.1, (i). Moreover, since the “abstract outer action pair” [i.e., an abstract profinite group out equipped with an outer action by another abstract profinite group] G 2  Δ 2 is out clearly isomorphic to the composite abstract outer action pair G 1 G 2  Δ 2 [as out well as, by definition, the abstract outer action pair G 1  Δ 1 ] i.e.,  out G 1  Δ 1   out G 1 G 2  Δ 2   out G 2  Δ 2  we thus conclude that the log-wall can be penetrated by the isomorphism class out of the abstract outer action pair G i  Δ i . G 1 log Δ 1 G 2 out out  log log  Δ 2 (ii) Once one has made the observations made in (i), it is natural to proceed to consider what sort of “additional data” may be shared on both sides of the log-wall. out Typically, “purely group-theoretic structures” constructed from “G i  Δ i serve as natural containers for such additional data [cf., e.g., the discussion of Remark 3.7.5]. Thus, the additional data may be thought as some sort of a choice [cf. the dotted arrows in the diagram below] among various possibilities [cf. the ’s” in the diagram below] housed in such a group-theoretic container. log log log log From this point of view: 100 SHINICHI MOCHIZUKI The fundamental difference that distinguishes the pseudo-mono-anabelian approach discussed in Remark 3.7.5 from the mono-anabelian approach is the issue of whether this “choice” is specified in terms that depend on the scheme theory that gives rise to the choice [i.e., the pseudo-mono- anabelian case] or not [i.e., the mono-anabelian case, in which the choice may be specified in language that depends only on the abstract group out structure of, say, “G i  Δ i ”]. In fact, the discussion in Remark 3.7.5, (ii) [cf. also Figs. 1, 2 of Remark 3.7.3, (iv)], may be depicted via a similar illustration to the above illustration of the “log- wall” in which the “log-wall” is replaced by a “model-wall” separating “reference models” from “log-subject copies” of such models. In Remark 3.7.5, (ii), special attention was given to the situation in which the “additional data” consists of the “additive structure” on the image of the Kummer map. When the Δ i ’s of (i) are given by the birational geometric fundamental groups “Δ η X of Theorem 1.11, another example of such “additional data” in which the specification of the “choice” depends on “scheme theory” [and hence cannot, at least a priori, be shared on both sides of the log-wall] is given by the specification of some finite collection of closed points corresponding to the cusps of some affine hyperbolic curve that lies in some given scheme theory [cf. Remark 1.11.5]. (iii) With regard to the issue of “specifying some finite collection of closed points corresponding to the cusps of an affine hyperbolic curve” discussed in the final portion of (ii), we note that in certain special cases, a “purely group-theoretic” specification is in fact possible. For instance, if, in the notation of Theorem 1.11, X is a hyperelliptic curve whose unique nontrivial k-automorphism is given by its hyperelliptic involution, then the set of points fixed by the hyperelliptic involution constitutes such an example in which a “purely group-theoretic” specification can be made by considering the conjugacy classes of inertia groups “I x fixed by the unique nontrivial outer automorphism of Δ η X that commutes with the given outer action of G k on Δ η X . (iv) The “log-wall” discussed in (i) is reminiscent of the constant indeterminacy arising from morphisms of Frobenius type [i.e., which thus constitute a “wall” that cannot be penetrated by constant rigidity] in the theory of the étale theta function [cf. [Mzk18], Corollary 5.12 and the following remarks], as well as of the subtleties that arise from the Frobenius morphism in the context of anabelian geometry in positive characteristic [cf., e.g., [Stix]]. Remark 3.7.8. Many of the arguments in the various remarks following Corollaries 3.6, 3.7 are not formulated entirely rigorously. Thus, in the future, it is quite possible that certain of the obstacles pointed out in these remarks can be overcome. Nevertheless, we presented these remarks in the hope that they could aid in elucidating the content of and motivation [from the point of view of the author] behind the various rigorously formulated results of the present paper. TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 101 Section 4: Archimedean Log-Frobenius Compatibility In the present §4, we present an archimedean version [cf. Corollary 4.5] of the theory of §3, i.e., we interpret the theory of §2 in terms of a certain compatibility with the “log-Frobenius functor”. Definition 4.1. (i) Let k be a CAF [cf. §0]. Write O k k for the subset of elements of absolute value 1, O k × O k for the subgroup of units [i.e., elements of absolute value equal to 1 cf. [Mzk17], Definition 3.1, (ii)], O k  O k for the multiplicative monoid of def nonzero elements, and k  k × = k\{0} for the universal covering of k × . Thus, k  k × is uniquely determined, up to unique isomorphism, as a pointed topological space and, moreover, [as is well-known] may be constructed explicitly by considering homotopy classes of paths on k × ; moreover, the pointed topological space k admits a natural topological group structure, determined by the topological group structure of k × . Note that the “inverse” of the exponential map k k × [given by the usual power series] determines an isomorphism of topological groups log k : k k which we shall refer to as the logarithm associated to k. Next, let X ell be an elliptically admissible Aut-holomorphic orbispace [cf. Definition 2.1, (i); Re- mark 2.1.1]. We shall refer to as a [k-]Kummer structure on X ell any isomorphism of topological fields  def κ k : k A X ell = A X ell {0} where we write A X ell for the “A p of Corollary 2.7, (e) [equipped with various topological and algebraic structures], which may be identified [hence considered as an object that is independent of “p”] via the various isomorphisms “A p A p  of Corollary 2.7, (e), together with the functoriality of the algorithms of Corollary 2.7. Note that k, k × , k , and A X ell are equipped with natural Aut-holomorphic structures, with respect to which κ k determines co-holomorphicizations between k and A X ell , as well as between k and A X ell ; moreover, these co-holomorphicizations are compatible with log k . Next, let T {TF, TCG, TLG, TM, TH, TH} where TF, TCG, TLG, TM as in Definition 3.1, (i), and we write TH for the category of connected Aut-holomorphic orbispaces and morphisms of Aut- holomorphic orbispaces [cf. Remark 2.1.1], and TH 102 SHINICHI MOCHIZUKI for the category of connected Aut-holomorphic groups [i.e., Aut-holomorphic spaces equipped with a topological group structure such that both the Aut-holomorphic and topological group structures arise from a single connected complex Lie group structure] and homomorphisms of Aut-holomorphic groups. If T is equal to TF (respectively, TCG; TLG; TM; TH; TH), then let M k Ob(T) be the object determined by k (respectively, the object determined by O k × ; the object deter- mined by k × ; the object determined by O k  ; any object of TH equipped with a co-holomorphicization κ M k : M k A X ell ; any object of TH equipped with an Aut-holomorphic homomorphism κ M k : M k A X ell (⊆ A X ell ) [relative to the mul- tiplicative structure of A X ell ]); if T  = TH, TH and κ k is a k-Kummer structure on X ell , then write κ M k : M k A X ell for the restriction of κ k to M k k. We shall refer to as a model Aut-holomorphic T-pair any collection of data (a), (b), (c) of the following form: (a) the elliptically admissible Aut-holomorphic orbispace X ell , (b) the object M k Ob(T), (c) the datum κ M k : M k A X ell . Also, we shall refer to the datum κ M k of (c) as the Kummer structure of the model κ Aut-holomorphic T-pair; we shall often use the abbreviated notation (X ell  M k ) for this collection of data (a), (b), (c). κ (ii) We shall refer to any collection of data (X  M ) consisting of an elliptically admissible Aut-holomorphic orbispace X, an object M Ob(T), and a datum κ κ M : M A X , which we shall refer to as the Kummer structure of (X  M ), as an Aut-holomorphic T-pair if the following conditions are satisfied: (a) κ M is a continuous map between the underlying topological spaces whenever T  = TH; (b) κ M is a collection of continuous maps from open subsets of the underlying topological space of M to the underlying topological space of A X whenever T = TH; κ (c) for some model Aut-holomorphic T-pair (X ell  M k ) [where the notation is as in (i)], there exist an isomorphism X ell X of objects of TH and an isomorphism M k M of objects of T that are compatible with the respective Kummer structures κ M k : M k A X ell , κ M : M A X . In this situation, we shall refer to X as the structure-orbispace and to M as the arithmetic data of the Aut-holomorphic T-pair κ (X  M ); if, in this situation, the structure-orbispace X arises from a hyperbolic orbicurve which is of strictly Belyi type [cf. Remark 2.8.3], then we shall refer to κ the Aut-holomorphic T-pair (X  M ) as being of strictly Belyi type. A morphism of Aut-holomorphic T-pairs κ κ φ : (X 1  M 1 ) (X 2  M 2 ) consists of a morphism of objects φ M : M 1 M 2 of T, together with a compatible [relative to the respective Kummer structures] finite étale morphism φ X : X 1 X 2 of TH; if, in this situation, φ M (respectively, φ X ) is an isomorphism, then we shall refer to φ as a T-isomorphism (respectively, structure-isomorphism). (iii) Write C T hol TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 103 for the category whose objects are the Aut-holomorphic T-pairs and whose mor- phisms are the morphisms of Aut-holomorphic T-pairs. Also, we shall use the same notation, except with “C” replaced by C (respectively, C; C) to denote the various subcategories determined by the T-isomorphisms (respec- tively, structure-isomorphisms; isomorphisms); we shall use the same notation, with “hol” replaced by hol-sB to denote the various full subcategories determined by the objects of strictly Belyi type. Since [in the notation of (i)] the formation of O k  (respectively, k × ; O k × ; O k × ) from k (respectively, O k  ; O k  ; k × ) is clearly intrinsically defined [i.e., depends only on the “input data of an object of T”], we thus obtain natural functors hol hol C TM ; C TF hol hol C TM C TLG ; hol hol C TM C TCG ; hol hol C TLG C TCG i.e., by taking the multiplicative monoid of nonzero elements of absolute value 1 of the arithmetic data [i.e., nonzero elements of the closure of the set of elements a such that a n 0 as n ∞], the associated groupification M gp of the arithmetic data M , the subgroup of invertible elements M × of the arithmetic data M , or the maximal compact subgroup of the arithmetic data. Finally, we shall write TH EA EA sB for the subcategories determined, respectively, by the elliptically admissible hyper- bolic orbicurves over CAF’s and the finite étale morphisms, and by the elliptically admissible hyperbolic orbicurves of strictly Belyi type over CAF’s and the finite étale morphisms; also, we shall use the same notation, except with “EA” replaced by EA to denote the subcategory determined by the isomorphisms. Thus, for κ T {TF, TCG, TLG, TM, TH, TH}, the assignment (X  M ) → X determines various compatible natural functors C T hol EA [as well as double underlined versions of these functors]. (iv) Observe that [in the notation of (i)] the field structure of k determines, via the inverse morphism to log k , a structure of topological field on the topological group k ; moreover, κ k determines a k -Kummer structure on X ell κ k : k A X ell which may be uniquely characterized [i.e., among the two k -Kummer structures on X ell ] by the property that the co-holomorphicization determined by κ k coincides with the co-holomorphicization determined by composing the composite of natural maps k  k × → k with the co-holomorphicization determined by κ k . In par- ticular, [cf. (i)] the co-holomorphicizations determined by κ k , κ k are compatible 104 SHINICHI MOCHIZUKI with log k . Since the various operations applied here to construct this field struc- ture on k [such as, for instance, the power series used to define log k ] are clearly intrinsically defined [cf. the natural functors defined in (iii)], we thus obtain that the construction that assigns (the topological field k, with its Kummer structure κ k ) → (the topological field k , with its Kummer structure κ k ) determines a natural functor hol hol C TF log TF,TF : C TF which we shall refer to as the log-Frobenius functor. Since log k determines a functorial isomorphism between the fields k, k , it follows immediately that the functor log TF,TF is isomorphic to the identity functor [hence, in particular, is an equivalence of categories]. By composing log TF,TF with the various natural functors defined in (iii), we also obtain, for T {TLG, TCG, TM}, a functor hol log TF,T : C TF C T hol which [by abuse of terminology] we shall also refer to as “the log-Frobenius functor”. In a similar vein, the assignments (the topological field k, with its Kummer structure κ k ) → (the Aut-holomorphic space k × , with its Kummer structure k | k × ]) (the topological field k, with its Kummer structure κ k ) → (the Aut-holomorphic space k , with its Kummer structure k ]) where the [−]’s denote the associated co-holomorphicizations; the phrase “the Aut-holomorphic space” should, strictly speaking, be interpreted as meaning “the Aut-holomorphic space determined by” determine natural functors hol hol C TH ; λ × : C TF hol hol λ : C TF C TH together with diagrams of functors hol C TF  λ hol C TH log TF,TF −→  hol C TF ×  λ hol C TF ι × λ ×    λ = hol C TH hol C TH ι log where we write ι log : λ × log TF,TF λ for the natural transformation in- duced by the natural inclusion “(k ) × → k and ι × : λ λ × for the natural transformation induced by the natural map “k  k × ”. Finally, we note that the fields “k obtained by the above construction [i.e., the arithmetic data of the objects in the image of the log-Frobenius functor log TF,TF ] are equipped with a nat- ural “subquotient compactum” i.e., the compact subset “O k × k × that lies in TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 105 the natural quotient “k  k × which we shall refer to as the pre-log-shell of κ log arith TF,TF ((X  M )) κ λ (X  M ) κ hol κ where (X  M ) Ob(C TF ); λ (X  is a subquotient of the arithmetic data M ) κ log arith TF,TF ((X  M )) of the object determined by applying the log-Frobenius functor κ log TF,TF to the object (X  M ). (v) Write LinHol [i.e., “linear holomorphic”] for the category whose objects are pairs   κ X, X  A X consisting of an object X Ob(EA), together with the tautological Kummer map A X A X [i.e., given by the identity on the object of TF determined by A X ] all of which is to be understood as constructed via the algorithms of Corollary 2.7 [cf. Remark 3.1.2] and whose morphisms are the morphisms induced by the [finite étale] morphisms of EA [cf. the functorial algorithms of Corollary 2.7]. Thus, we obtain a natural functor κ LH EA −→ LinHol which [as is easily verified] is an equivalence of categories, a quasi-inverse for which is given by the natural projection functor LinHol EA. Remark 4.1.1. The topological monoid “O k  associated to a CAF k [cf. Defini- tion 4.1, (i)] is essentially the data used to construct the archimedean Frobenioids of [Mzk17], Example 3.3, (ii). Remark 4.1.2. Although, to simplify the discussion, we have chosen to require that the structure-orbispace always be elliptically admissible, and that the base field always be a CAF, many aspects of the theory of the present §4 may be general- ized to accommodate “structure-orbispaces” that are Aut-holomorphic orbispaces that arise from more general hyperbolic orbicurves [cf., e.g., Propositions 2.5, 2.6; Remark 2.6.1] over arbitrary archimedean fields [i.e., either CAF’s or RAF’s cf. §0]. Such generalizations, however, are beyond the scope of the present paper. Proposition 4.2. (First Properties of Aut-Holomorphic Pairs) κ κ (i) Let T {TM, TF, TLG, TCG}; (X  M ), (X  M ) Ob(C T hol ). Then the natural functor of Definition 4.1, (iii), induces a bijection [cf. Proposition 3.2, (iv)] κ κ Isom C hol ((X  M ), (X  M )) Isom EA (X, X ) T hol-sB = on sets of isomorphisms. In particular, the categories EA, C T hol = C hol T , C T hol-sB are id-rigid. C T (ii) The equivalence of categories κ LH : EA LinHol of Definition 4.1, (v) i.e., the functorial algorithms of Corollary 2.7 determines a natural [1- ]factorization [cf. Proposition 3.2, (v)] hol C TF −→ hol C TM log TM,T −→ C T hol 106 SHINICHI MOCHIZUKI where T {TF, TLG, TCG, TM}; the first arrow is the natural functor of Defini- hol C T hol of Definition tion 4.1, (iii) of the log-Frobenius functors log TF,T : C TF 4.1, (iv). Moreover, [when T {TF, TM}] the functor log T,T is isomorphic to the identity functor [hence, in particular, is an equivalence of categories]. Proof. The bijectivity portion of assertion (i) follows immediately from the required compatibility of morphisms of C T hol with the Kummer structures of the objects in- volved [cf. also the functorial algorithms of Corollary 2.7]. To verify the id-rigidity of EA, it suffices to observe that for any object X Ob(EA), which necessarily arises from some elliptically admissible hyperbolic orbicurve X over a CAF, the full subcategory of EA consisting of objects that map to X may, by Corollary 2.3, (i) [cf. also [Mzk14], Lemma 1.3, (iii)], be identified with the category of finite étale R-localizations “Loc R (X)” [i.e., the subcategory of the category of finite étale localizations “Loc(X)” of [Mzk10], §2, obtained by considering the R-linear mor- phisms]. Thus, the id-rigidity of EA follows immediately from the slimness assertion of Lemma 4.3 below. In light of the bijectivity portion of assertion (i), the id-rigidity hol-sB = C hol-sB follows in a similar fashion. This of the categories C T hol = C hol T , C T T completes the proof of assertion (i). Assertion (ii) follows immediately from the definitions [and the functorial algorithms of Corollary 2.7]. Remark 4.2.1. Note that, unlike the case with Proposition 3.2, (iv), the id- rigidity portion of Proposition 4.2, (i), is [as is easily verifed] false for the “C” and “C” versions of C T hol , C T hol-sB . The following result is well-known. Lemma 4.3. (Slimness of Archimedean Fundamental Groups) Let X be a hyperbolic orbicurve over an archimedean field k X . Then the étale fundamental group Π X of X is slim. Proof. Let k X be an algebraic closure of k X . Thus, we have an exact sequence of profinite groups 1 Δ X Π X G 1 def def [where Δ X = π 1 (X × k X k X ); G = Gal(k X /k X )]. Since Δ X is slim [cf., e.g., [Mzk20], Proposition 2.3, (i)], it suffices to consider the case where there exists an element σ Π X that maps to a nontrivial element σ G G = Z/2Z and, moreover, commutes with some open subgroup H Π X . We may assume without loss of generality that H Δ X , and, moreover, that H corresponds to a finite étale covering of X × k X k X which is a hyperbolic curve of genus 2. In particular, by replacing X by the finite étale covering of X determined by the open subgroup of Π X generated by H and σ, we may assume that σ lies in the center of Π X , and, moreover, that X is a hyperbolic curve of genus 2. In particular, by filling in the cusps of X, we may assume further that X is proper. Now if l is any prime number, then the first Chern class of, say, the canonical bundle of X determines a generator of H 2 (X × k X k X , Q l (1)) = H 2 X , Q l (1)) [where the “(1)” denotes a Tate twist], hence an isomorphism of G-modules H 2 X , Q l ) Q l (−1). In particular, TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 107 it follows that σ acts nontrivially on H 2 X , Q l ), in contradiction to the fact that σ lies in the center of Π X . This contradiction completes the proof of Lemma 4.3. Lemma 4.4. (Topological Distinguishability of Additive and Multi- plicative Structures) Let k be a CAF [cf. Definition 4.1, (i)]. Then [in the notation of Definition 4.1, (i)] no composite of the form k × α −→ (k ) × → k  k × where the “×” of “(k ) × is relative to the field structure of k [cf. Definition 4.1, (iv)]; α is an isomorphism of topological groups; “→” is the natural inclusion; “” is the natural map is bijective. Proof. Indeed, the non-injectivity of k  k × implies that the composite under consideration fails to be injective. Corollary 4.5. (Aut-Holomorphic Mono-anabelian Log-Frobenius Com- patibility) Write def def hol X = C hol T = C T ; E = EA; def hol N = C TH where [in the notation of Definition 3.1] T {TM, TF}. Consider the diagram of categories D ... ... X log −→ X id   id +1  log −→ X ...  id −1 ... X λ ×   λ N  E κ LH  LinHol  E where we use the notation “log”, “λ × ”, “λ for the arrows “log T,T ”, “λ × ”, “λ of Definition 4.1, (iv) [cf. also Proposition 4.2, (ii)]; we employ the conven- tions of Corollary 3.6 concerning subdiagrams of D; we write L for the countably ordered set determined by [cf. §0] the infinite linear oriented graph  Γ opp D ≤1 [so the elements of L correspond to vertices of the first row of D] and L = L {} def 108 SHINICHI MOCHIZUKI for the ordered set obtained by appending to L a formal symbol  [which we think of as corresponding to the unique vertex of the second row of D] such that  < , for all  L; id  denotes the identity functor at the vertex  L. Then: (i) For n = 4, 5, 6, D ≤n admits a natural structure of core on D ≤n−1 . That is to say, loosely speaking, E, LinHol “form cores” of the functors in D. (ii) The assignments    κ κ X, X  A X → (X  A X ), κ (X  A X )  [where we write “A for the monoid of nonzero elements of absolute value 1 of the CAF given by “A”] determine [i.e., for each choice of T] a natural “forgetful” functor φ LH LinHol −→ X which is an equivalence of categories, a quasi-inverse for which is given by the composite π LH : X LinHol of the natural projection functor X E with κ LH : E LinHol; write η LH : φ LH π LH id X for the tautological isomorphism arising from the definitions [cf. Definition 4.1, (i), (ii)]. Moreover, φ LH gives rise to a telecore structure T LH on D ≤4 , whose underlying diagram of categories we denote by D LH , by appending to D ≤5 telecore edges ... φ +1 ... X LinHol φ    log −→ LinHol  φ −1 ... X ... log X −→ φ  −→ X from the core LinHol to the various copies of X in D ≤2 given by copies of φ LH , which we denote by φ  , for  L . That is to say, loosely speaking, φ LH determines 0 ] for the path a telecore structure on D ≤4 . Finally, for each  L , let us write  1   on Γ D LH of length 0 at  and  ] for the path on Γ D LH of length {4, 5} [i.e., depending on whether or not  = ] that starts from , descends [say, via λ × ] to the core vertex “LinHol”, and returns to  via the telecore edge φ  . Then the collection of natural transformations −1 −1  , η  , η  , η  } ∈L,∈L where we write η  for the identity natural transformation from the arrow φ  : LinHol X to the composite arrow id  φ  : LinHol X and η  : (D LH )  1 ] (D LH )  0 ] for the isomorphism arising from η LH generate a contact structure H LH on the telecore T LH . (iii) The natural transformations ι log, : λ × id  log λ id +1 , ι × : λ λ × TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 109 [cf. Definition 4.1, (iv)] belong to a family of homotopies on D ≤3 that determines on D ≤3 a structure of observable S log on D ≤2 and, moreover, is compatible with the families of homotopies that constitute the core and telecore structures of (i), (ii). (iv) The diagram of categories D ≤2 does not admit a structure of core on D ≤1 which [i.e., whose constituent family of homotopies] is compatible with [the constituent family of homotopies of ] the observable S log of (iii). Moreover, the telecore structure T LH of (ii), the contact structure H LH of (ii), and the observable S log of (iii) are not simultaneously compatible [but cf. Remark 3.7.3, (ii)]. (v) The unique vertex  of the second row of D is a nexus of  Γ D . More- over, D is totally -rigid, and the natural action of Z on the infinite linear oriented graph  Γ D ≤1 extends to an action of Z on D by nexus-classes of self- equivalences of D. Finally, the self-equivalences in these nexus-classes are com- patible with the families of homotopies that constitute the cores and observ- able of (i), (iii); these self-equivalences also extend naturally [cf. the technique of extension applied in Definition 3.5, (vi)] to the diagram of categories [cf. Definition 3.5, (iv), (a)] that constitutes the telecore of (ii), in a fashion that is compatible with both the family of homotopies that constitutes this telecore structure [cf. Definition 3.5, (iv), (b)] and the contact structure H LH of (ii). Proof. Assertions (i), (ii) are immediate from the definitions [and the functorial algorithms of Corollary 2.7] cf. also the proofs of Corollary 3.6, (i), (ii). Next, × we consider assertion (iii). If, for  L, one denotes by “k  the arithmetic data of type TLG [which we may be obtained from an arithmetic data of type T {TM, TF} via the natural functors of Definition 4.1, (iii)] of a “typical object” of the copy of X at the vertex  of D ≤1 , then ι × “applied at the vertex ” corresponds × to the natural surjection k   k  , while ι log, corresponds to the natural inclusion × × × k  → k +1 , where we think of k  as being obtained from k +1 via the application of log. In particular, by letting  L vary and composing these natural surjections and inclusions, we obtain a diagram ... × × ×  k  → k +1  k +1 → k +2  k +2 → → k  ... [which is compatible with the various Kummer structures cf. Remark 4.5.1, (i), hol in Definition 4.1, (i), (ii), (iii)]. The paths on [the below; the definition of C TH oriented graph corresponding to] this diagram may be classified into four types, which correspond [by composing, in an alternating fashion, various pull-backs of “ι log, with various pull-backs of ι × ] to homotopies on D ≤3 , as follows [cf. the notational conventions of the proof of Corollary 3.6]: × (1) the path corresponding to the composite “k  k +n ”, which yields a × n homotopy for pairs of paths ([λ ] [id  ] [log] [γ], ] [id +n ] [γ]) × × (2) the path corresponding to the composite “k  k +n ”, which yields a × n × homotopy for pairs of paths ([λ ] [id  ] [log] [γ], ] [id +n ] [γ]) k +n ”, which yields a (3) the path corresponding to the composite “k  n homotopy for pairs of paths ([λ ] [id  ] [log] [γ], ] [id +n ] [γ]) 110 SHINICHI MOCHIZUKI × (4) the path corresponding to the composite “k  k +n−1 ”, which yields a n−1 × homotopy for pairs of paths ([λ ]◦[id  ]◦[log] ◦[γ], ]◦[id +n−1 ]◦[γ]) where n 1 is an integer and [γ] is a path on D ≤1 ; in the case of type (4), it is convenient to include also the pair of paths ([λ ], × ]), for which the natural transformation ι × determines a homotopy. In addition, it is natural to consider the “identity homotopies” associated to the pairs (5) ([γ], [γ]), where [γ] is a path on D ≤3 whose terminal vertex lies in the third row of D ≤3 . Thus, if we take E log to be the set of ordered pairs of paths on  Γ D ≤3 consisting of pairs of paths of the above five types, then one verifies immediately that E log satisfies the conditions (a), (b), (c), (d), (e) given in §0 for a saturated set. In particular, the various homotopies discussed above yield a family of homotopies which determines an observable S log , as desired. Moreover, it is immediate from the definitions i.e., in essence, because the various structure-orbispaces that appear remain “undisturbed” by the various manipulations involving arithmetic data that arise from “ι log, ”, “ι × that this family of homotopies is compatible with the families of homotopies that constitute the core and telecore structures of (i), (ii). This completes the proof of assertion (iii). Next, we consider assertion (iv). Suppose that D ≤2 admits a structure of core on D ≤1 in a fashion that is compatible with the observable S log of (iii). Then this core structure determines a homotopy ζ 0 for the pair of paths ([id  ], [id −1 ] [log]) [for  L]; thus, by composing the result ζ 0  of applying λ × to ζ 0 with the homotopy ζ 1 associated [via S log ] to the pair of paths ([λ × ] [id −1 ] [log], ] [id  ]) [of type (1)] and then with the homotopy ζ 2 associated [via S log ] to the pair of paths ([λ ] [id  ], × ] [id  ]) [of type (4)], we obtain a natural transformation ζ 1  = ζ 2 ζ 1 ζ 0  : λ × id  λ × id  which, in order for the desired compatibility to hold, must coincide with the “identity homotopy” [of type (5)]. On the other hand, by writing out explicitly the meaning of such an equality ζ 1  = id, we conclude that we obtain a contradiction to Lemma 4.4. This completes the proof of the first incompatibility of assertion (iv). The proof of the second incompatibility of assertion (iv) is entirely similar [cf. the proof of Corollary 3.6, (iv)]. This completes the proof of assertion (iv). Finally, the total -rigidity portion of assertion (v) follows immediately from Proposition 4.2, (i) [cf. also the final portion of Proposition 4.2, (ii)]; the remainder of assertion (v) follows immediately from the definitions. Remark 4.5.1. (i) The “output” of the observable S log of Corollary 4.5, (iii), may be summa- rized intuitively in the following diagram [cf. Remark 3.6.1, (i)]: ... × k +1 k +1 κ  κ  ... X +1 ← X +1 × k  k  κ  κ  X  ← X  × k −1 k −1 κ  κ  X −1 ... X −1 ... TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 111 × where the arrows are the natural surjections [cf. ι × !]; k  , for  L, is a × copy of “k that arises, via id  , from the vertex  of D ≤1 ; the arrows “←” are the × is obtained by applying the log-Frobenius inclusions arising from the fact that k  κ × functor log to k +1 [cf. ι log, !]; the  ’s” denote the various Kummer structures involved; the isomorphic “X  ’s” correspond to the coricity of E [cf. Corollary 4.5, (i)]. Finally, the incompatibility assertions of Corollary 4.5, (iv), may be thought of as a statement of the non-existence of some “universal reference model” × k model × that maps isomorphically to the various k  ’s in a fashion that is compatible with the various arrows ”, “←” of the above diagram. (ii) In words, the essential content of Corollary 4.5 may be understood as follows [cf. the “intuitive diagram” of (i)]: Although the operation represented by the log-Frobenius functor is com- patible with the [Aut-holomorphic] structure-orbispaces, hence with the “software” constituted by the algorithms of Corollary 2.7, it is not compat- ible with the additive or multiplicative structures on the various arithmetic data involved cf. Remark 3.6.1. That is to say, more concretely, if one starts with an elliptically admissible Aut- holomorphic orbispace X on which [for some CAF k] k × “acts via the local linear holomorphic structures of Corollary 2.7, (e)” [i.e., X is equipped with a Kummer κ structure X  k × ], then applies log k to the universal covering k k × to equip k with a field structure, with respect to which k “acts” on some isomorph X  of X k ← (k ) × k × κ κ  X  X  [where the “×” of “(k ) × is taken with respect to this field structure of k ], then although the “actions” of k × , (k ) × on X X  are not strictly compatible [i.e., the diagram does not commute], they become “compatible” if one “loosens one’s notion of compatibility” to the notion of being “compatible with the [Aut-]holomorphic structure” of the various objects involved [cf. the analogy of Remark 2.7.3]. This state of affairs may be expressed formally as a compatibility between the various hol in Definition 4.1, (i), co-holomorphicizations involved [cf. the definition of C TH (ii), (iii)]. In summary, as should be evident from its statement, Corollary 4.5 is intended as an archimedean analogue of Corollary 3.6. In particular, the “general formal content” of Remarks 3.6.1, 3.6.2, 3.6.3, 3.6.5, 3.6.6, and 3.6.7 applies to the present archimedean situation, as well. Remark 4.5.2. By comparison to the nonarchimedean case treated in §3, certain but not all! of the “arrows” that appear in the archimedean case go in the opposite direction to the nonarchimedean case. This is somewhat reminiscent of the “product formula” in elementary number theory, where, for instance, positive 112 SHINICHI MOCHIZUKI powers of prime numbers 0 at nonarchimedean primes, but at archimedean primes. In the context of Corollary 4.5, perhaps the most important example of this phenomenon is given by “ι × ”. This leads to a somewhat different structure for the observable S log of Corollary 4.5, (iii) involving “archimedean” homotopies of arbitrarily large “length” [cf. the “non-[γ]-portion” of the pairs of paths of types (1), (2), (3), (4) in the proof of Corollary 4.5, (iii)] from the structure of the observable S log of Corollary 3.6, (iii) which involves “nonarchimedean” paths of bounded “length” [cf. the “non-[γ]-portion” of the pairs of paths of types (1), (2) in the proof of Corollary 3.6, (iii)]. Remark 4.5.3. (i) By replacing “λ ×pf by “λ ”, “ι × = ι × : λ × λ ×pf by “ι × = ι × : λ λ × ”, and “Corollary 1.10” by “Corollary 2.7”, [and making various other suitable revisions] one obtains an essentially straightfor- ward “Aut-holomorphic translation” of the bi-anabelian incompatibility result given in Corollary 3.7. We leave the routine details to the reader. (ii) The “general formal content” of Remarks 3.7.1, 3.7.2, 3.7.3, 3.7.4, 3.7.5, 3.7.7, and 3.7.8 applies to the archimedean analogue of Corollary 3.7 discussed in (i) cf. also the analogy of Remark 2.7.3; the discussion in Remark 2.7.4 of “fixed reference models” in the context of the definition of the notion of a “holomorphic structure”. (iii) With regard to the discussion in Remark 3.7.4 of “functorially trivial mod- els” [i.e., models that “arise from Π” without essential use of Π, hence are equipped with trivial functorial actions of Π], we note that although “the Galois group Π” does not appear in the present archimedean context, the “functorial detachment” of such “functorially trivial models” means, for instance, that if one regards some model X model as “arising” from an elliptically admissible Aut-holomorphic orbis- pace X in a “trivial fashion”, then when one applies the “elliptic cuspidalization” portion of the algorithm of Corollary 2.7, (b), the various coverings of X involved in this elliptic cuspidalization algorithm functorially induce trivial coverings of X model , hence do not give rise to a functorial isomorphism of the respective “base fields” [cf. Remark 2.7.3] of X, X model . (iv) With regard to the discussion in Remark 3.7.5, one may give an archimedean analogue of the “pathological versions of the Kummer map” given in Remark 3.7.5, (ii), by composing the k-Kummer structure [cf. Definition 4.1, (i)] “κ k : k A X ell ”, restricted, say, to k × , with the [non-additive!] automorphism of k × O k × × R >0 that acts as the identity on O k × and is given by raising to the λ-th power [for some λ R >0 ] on R >0 . TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 113 Section 5: Global Log-Frobenius Compatibility In the present §5, we globalize the theory of §3, §4. This globalization allows one to construct canonical rigid compacta i.e., canonical integral structures that enable one to consider [“pana-”]localizations of global arithmetic line bundles [cf. Corollary 5.5] without obliterating the “volume-theoretic” information inherent in the theory of global arithmetic degrees, and in a fashion that is compatible with the operation of “mono-analyticization” [cf. Corollary 5.10] i.e., the operation of “disabling the rigidity” of one of the “two combinatorial dimensions” of a ring [cf. Remark 5.6.1]. The resulting theory is reminiscent, in certain formal respects, of the p-adic Teichmüller theory of [Mzk1], [Mzk4] [cf. Remark 5.10.3]. Definition 5.1. (i) Let F be a number field. Then we shall write V(F ) for the set of [archimedean def { F }, where the and nonarchimedean] valuations of F , and V  (F ) = V(F ) symbol F is to be thought of as representing the global field F , or, alternatively, the generic prime of F . If F is an algebraic closure of F , then we shall write def V(F /F ) = lim ←− V(K); K  V  (F /F ) = lim ←− V (K) def K [where K ranges over the finite extensions of F in F ] for the inverse limits relative to the evident systems of morphisms. The inverse system of K ’s” determines a unique global element F V  (F /F ); the other elements of V  (F /F ) lie in the image of the natural injection V(F /F ) → V  (F /F ) and will be called local; moreover, we have a natural decomposition  V(F /F ) = V(F /F ) arc V(F /F ) non into archimedean and nonarchimedean local elements. There is a natural contin- uous action of Gal(F /F ) on the pro-sets V(F /F ), V  (F /F ). For K F a fi- nite extension of F , V(K), V  (K) may be identified, respectively, with the sets of Gal(F /K) (⊆ Gal(F /F ))-orbits V(F /F )/Gal(F /K), V  (F /F )/Gal(F /K) of V(F /F ), V  (F /F ). (ii) Let X be an elliptically admissible [cf. [Mzk21], Definition 3.1] hyperbolic orbicurve over a totally imaginary number field F [so X is also of strictly Belyi type cf. Remark 2.8.3]. Write Π X for the étale fundamental group of X [for def some choice of basepoint]; Π X  G F = Gal(F /F ) for the natural surjection onto the absolute Galois group G F of F [for some choice of algebraic closure F of F ]; Δ X Π X for the kernel of this surjection [which may be characterized “group- theoretically” as the maximal topologically finite generated closed normal subgroup of Π X cf., e.g., [Mzk9], Lemma 1.1.4, (i)]. Write F mod F for the “field of moduli of X”, i.e., the subfield of F determined by the [open] image def of Aut(X F ) [i.e., the group of automorphisms of the scheme X F = X × F F ] in 114 SHINICHI MOCHIZUKI Aut(F ) = Gal(F /Q) (⊇ G F ); Aut(X), Aut(F ) for the respective automorphism groups of the schemes X, Spec(F ). For simplicity, we also make the following assumption on X: F is Galois over F mod ; the natural homomorphism Aut(X) Aut(F ) surjects onto Gal(F/F mod ) (⊆ Aut(F )); we have a natural isomorphism Aut(X/F ) Aut(X F /F ) between the group of F -automorphisms of X and the group of F -automor- phisms of X F . This assumption on X implies that we have a natural isomorphism Aut(X) × Gal(F/F mod ) Gal(F /F mod ) Aut(X F ) [induced by the fiber product structure X F = X × F F ], and hence that the natural exact sequence 1 Aut(X F /F ) Aut(X F ) Gal(F /F mod ) 1 admits a natural surjection onto the natural exact sequence 1 Aut(X/F ) Aut(X) Gal(F/F mod ) 1 [induced by the projection Aut(X) × Gal(F/F mod ) Gal(F /F mod )  Aut(X) to the first factor]. Note that by the functoriality of the algorithms of Theorem 1.9, it fol- lows that there is a natural isomorphism Aut(X) Out(Π X ) that is compatible with the natural morphisms Aut(X)  Gal(F/F mod ) → Aut(F ) = Out(G F ) [cf., e.g., [Mzk15], Theorem 3.1], Out(Π X ) Out(G F ); in particular, one may functorially construct the image G F mod → Aut(G F ) as the inverse image [i.e., via the natural projection Aut(G F ) Out(G F )] of the image of Out(Π X ) Out(G F ). Next, ob- × serve that one may functorially construct “F from Π X as the field “k NF {0} ( = k NF )” constructed in Theorem 1.9, (e) [cf. also Remark 1.10.1, (i)]; denote this field × constructed from Π X by k NF X ); we shall also use the notation k NF X ) for the group of nonzero elements of this field. In particular, by considering [cf. Corollary 2.8] valuations on the field k NF X ) [where each valuation is valued in the “copy of × R” given by completing the group “k NF with respect to the “order topology” de- termined by the valuation], one may functorially construct “V  (F /F )”, “V(F /F )” from Π X ; denote the resulting pro-sets constructed in this way by V  X ), V(Π X ) and the completion of k NF X ) at v V(Π X ) by k NF X , v). For v V(F /F ) non , write Π X,v Π X for the decomposition group of v [i.e., the closed subgroup of elements of Π X that fix v]; for v V(F /F ) arc , write X ell,v for the Aut-holomorphic orbispace “X v [associated to X at v] of Corollary 2.8, δ ell,v : Δ X π 1 (X ell,v ) for the natural outer isomorphism of Δ X with the profinite completion [denoted by the superscript “∧”] of the topological fundamental group of X ell,v , and κ ell,v : k NF X ) → A X ell,v TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 115 for the natural inclusion of fields [i.e., arising from the isomorphism of topological fields of Corollary 2.9, (b)]. When we wish to regard X ell,v as an object constructed from Π X [cf. Corollary 2.8], we shall use the notation X(Π X , v) [where we regard v as an element of V(Π X ) arc ]. Finally, we observe that Aut(Π X ) acts naturally on all of these objects constructed from Π X . In particular, we have a natural bijection def V  X )/Aut(Π X ) V  (F mod ). For v V(F mod ), write d mod = [F v F : (F mod ) v ] v for the degree of the completion of F at any v F V(F ) that divides v over the completion (F mod ) v of F mod at v. (iii) Write EA  for the category whose objects are profinite groups isomorphic to Π X for some X as in (ii), and whose morphisms are open injections of profinite groups that in- duce isomorphisms between the respective maximal topologically finitely generated closed normal subgroups [i.e., the respective “Δ X ”]. We shall refer to as a global Galois-theater any collection of data  V  =  V , v } v∈V non , {(X v , δ v , κ v )} v∈V arc ) def where Π Ob(EA  ); we shall refer to Π as the global Galois group of the Galois- theater; we write Δ Π for the maximal topologically finitely generated closed  normal subgroup of Π; V is a pro-set equipped with a continuous action by Π that  non arc non arc def decomposes into a disjoint union V = { V } V V V = V V ; non arc for v V , Π v Π is the closed subgroup of elements that fix v; for v V , X v is an Aut-holomorphic orbispace, δ v : Δ π 1 (X v ) is an outer isomorphism of profinite groups, and κ v : k NF (Π) → A X v is an inclusion of fields such that there exists a(n) [unique! cf. Remark 5.1.1 below] isomorphism of pro-sets ψ V : V  (Π) V  which we shall refer to as a reference isomorphism for V  that satisfies the fol- lowing conditions: (a) ψ V is Π-equivariant and maps k NF (Π) → V , V  (Π) non non arc arc V , V  (Π) arc V ; (b) for V  (Π) arc v ell → v V , there exists a(n) [unique! cf. Remark 5.1.1 below] isomorphism ψ v : X(Π, v ell ) X v of Aut- holomorphic spaces that is compatible with δ ell,v ell , δ v , as well as with κ ell,v ell , κ v . A morphism of global Galois-theaters  φ : 1  V 1 ,{(Π 1 ) v 1 }, {((X 1 ) v 1 , δ v 1 , κ v 1 )})  2  V 2 , {(Π 2 ) v 2 }, {((X 2 ) v 2 , δ v 2 , κ v 2 )}) is defined to consist of a morphism φ Π : Π 1 → Π 2 of EA  and a(n) [uniquely   determined cf. Remark 5.1.1 below] isomorphism of pro-sets φ V : V 1 V 2 that satisfy the following conditions: (a) φ Π , φ V are compatible with the actions   non non arc arc of Π 1 , Π 2 on V 1 , V 2 , and map V 1 → V 2 , V 1 V 2 , V 1 V 2 ; (b) arc arc for V 1 v 1 → v 2 V 2 , there exists a [unique! cf. Remark 5.1.1 below] isomorphism φ v : X v 1 X v 2 of Aut-holomorphic spaces that is compatible with 116 SHINICHI MOCHIZUKI non δ v 1 , δ v 2 , as well as with κ v 1 , κ v 2 . [Here, we note that (a) implies that for V 1 non v 1 → v 2 V 2 , φ Π induces an open injection φ v : Π v 1 → Π v 2 .] (iv) In the notation of (iii), we shall refer to as a panalocal Galois-theater any collection of data def V  = (V  , v } v∈V non , {X v } v∈V arc ) where V  is a set that decomposes as a disjoint union V  = { V } V non V arc def V = V non V arc ; for v V non , Π v Ob(Orb(TG)) [cf. §0; Definition 3.1, (iii)]; for v V arc , X v Ob(Orb(EA)) [cf. Definition 4.1, (iii)] such that there exists a Π Ob(EA  ) and an isomorphism of sets ψ V : V  (Π)/Aut(Π) V  which we shall refer to as a reference isomorphism for V  that satisfies the following conditions: (a) the composite of ψ V with the quotient map V  (Π)  V  (Π)/Aut(Π) maps k NF (Π) → V , V(Π) non  V non , V(Π) arc  V arc ; (b) for each v V non , Π v is isomorphic to the object of Orb(TG) determined by “the decomposition group Π v Π of v, considered up to automorphisms of Π v , as v V(Π) ranges over the elements lying over v”; (c) for each v V arc , X v is isomorphic to the object of Orb(EA) determined by “the Aut-holomorphic orbispace X(Π, v), considered up to automorphisms of X(Π, v), as v V(Π) ranges over the elements lying over v”. A morphism of panalocal Galois-theaters φ : (V 1  , {(Π 1 ) v 1 }, {(X 1 ) v 1 }) (V 2  , {(Π 2 ) v 2 }, {(X 2 ) v 2 }) is defined to consist of a bijection of sets φ V : V 1  V 2  that induces bijections V 1 non V 2 non , V 1 arc V 2 arc , together with open injections of [orbi-]profinite groups 1 ) v 1 → 2 ) v 2 [where V 1 non v 1 → v 2 V 2 non ; we recall that, in the notation of (ii), “F/F mod is Galois], and isomorphisms of [orbi-]Aut-holomorphic orbis- paces (X 1 ) v 1 (X 2 ) v 2 [where V 1 arc v 1 → v 2 V 2 arc ]. [Here, we observe that the existence of the isomorphisms “(X 1 ) v 1 (X 2 ) v 2 implies by considering Euler characteristics [cf. also [Mzk20], Theorem 2.6, (v)] that the open injec- tions “(Π 1 ) v 1 → 2 ) v 2 induce isomorphisms “Δ 1 Δ 2 between the respective geometric fundamental groups.] Write Th  (respectively, Th  ) for the category of global (respectively, panalocal) Galois-theaters and morphisms of global (respec- tively, panalocal) Galois-theaters. Thus, it follows immediately from the definitions that we obtain a natural “panalocalization functor” Th  Th  which is essentially surjective. (v) Let T {TF, TM, TLG} [cf. the notation of Definition 3.1, (i)]. If T = TF, def def then let T  = T; if T  = TF, then let T  = TLG; if T   = T, then a superscript “T  will be used to denote the operation of groupification of a monoid [i.e., “gp”]; if T  = T, then a superscript “T  will be used to denote the “identity operation” [i.e., may be ignored]. If Π Ob(EA  ), then let us write M T  (Π) for the object of T  , equipped with a continuous action by Π, determined by k NF (Π) [if T  = TF], TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 117 × k NF (Π) [if T  = TLG], equipped with the discrete topology; if v V(Π), then let us write M T (Π, v) for the object of T, equipped with a continuous action by the × decomposition group Π v Π of v, determined by k NF (Π, v) [if T  = TF], k NF (Π, v) [if T  = TM]. A global T-pair is defined to be a collection [if T  = TLG], O  k NF (Π,v) of data κ M  = (V  , M  , v } v∈V , {(Π v  M v )} v∈V non , {(X v  M v )} v∈V arc ) def where  V  =  V , v } v∈V non , {(X v , δ v , κ v )} v∈V arc ) non arc V ; M  Ob(T  ), which we shall refer is a global Galois-theater; V = V  to as the global arithmetic data of M , is equipped with a continuous action by non Π; for each v V , v  M v ) is an MLF-Galois T-pair with Galois group arc κ given by Π v ; for each v V , (X v  M v ) is an Aut-holomorphic T-pair with  structure-orbispace given by X v ; for each v V , ρ v : M  M v T is a [“re- striction”] morphism in T  such that, relative to some reference isomorphism  ψ V : V  (Π) V for V  as in (iii), there exist isomorphisms [in T  , T, respec- tively] ψ  : M T  (Π) M  ; v : M T (Π, v) M v } v∈V which we shall refer to as reference isomorphisms for M  that satisfy the non , ψ v is Π v -equivariant; following conditions: (a) ψ  is Π-equivariant; (b) for v V arc κ (c) for v V , the composite of ψ v with the Kummer structure of (X v  M v ) is compatible with κ v ; (d) ψ  , v } v∈V are compatible with the v } v∈V , relative  to the natural restriction morphisms ρ v (Π) : M T  (Π) M T (Π, v) T . In this situation, if T  = TF, then we shall refer to the profinite Π-module μ  (M  ) = Hom(Q/Z, M  ) Z def  as the cyclotome associated to this global T-pair and [which is isomorphic to Z] def (M  ) Q/Z. A morphism of global T-pairs write μ Q/Z (M  ) = μ  Z κ φ : (V 1  ,M 1  , v 1 }, {((Π 1 ) v 1  (M 1 ) v 1 )}, {((X 1 ) v 1  (M 1 ) v 1 )}) κ (V 2  , M 2  , v 2 }, {((Π 2 ) v 2  (M 2 ) v 2 )}, {((X 2 ) v 2  (M 2 ) v 2 )}) is defined to consist of a morphism of global Galois-theaters φ V  : V 1  V 2  , together with an isomorphism φ  : M 1  M 2  of T  , and isomorphisms φ v 1 : v 1 → v 2 V 2 ] in T, that satisfy the following (M 1 ) v 1 (M 2 ) v 2 [where V 1 compatibility conditions: (a) φ  is equivariant with respect to the open injection non Π 1 → Π 2 arising from φ V  ; (b) for v 1 V 1 , the isomorphism φ v 1 is compatible with the actions of 1 ) v 1 , 2 ) v 2 , relative to the open injection 1 ) v 1 → 2 ) v 2 arc induced by φ V  ; (c) for v 1 V 1 , the isomorphism φ v 1 is compatible with the κ κ Kummer structures of ((X 1 ) v 1  (M 1 ) v 1 ), ((X 2 ) v 2  (M 2 ) v 2 ), relative to the iso- morphism (X 1 ) v 1 (X 2 ) v 2 induced by φ V  ; (d) φ  , v 1 } v 1 ∈V 1 are compatible with the v 1 } v 1 ∈V 1 , v 2 } v 2 ∈V 2 . 118 SHINICHI MOCHIZUKI (vi) In the notation of (v), a panalocal T-pair is defined to be a collection of data κ M  = (V  , {(Π v  M v )} v∈V non , {(X v  M v )} v∈V arc ) def where V  = (V  , v } v∈V non , {X v } v∈V arc ) is a panalocal Galois-theater; for each v V non , v  M v ) is a(n) [strictly speaking, “orbi-”]MLF-Galois T-pair with κ Galois group given by Π v ; for each v V arc , (X v  M v ) is a(n) [strictly speaking, “orbi-”]Aut-holomorphic T-pair with structure-orbispace given by X v . A morphism of panalocal T-pairs κ φ : (V 1  ,{((Π 1 ) v 1  (M 1 ) v 1 )}, {((X 1 ) v 1  (M 1 ) v 1 )}) κ (V 2  , {((Π 2 ) v 2  (M 2 ) v 2 )}, {((X 2 ) v 2  (M 2 ) v 2 )}) is defined to consist of a morphism of panalocal Galois-theaters φ V  : V 1  V 2  , together with compatible T-isomorphisms of [orbi-]MLF-Galois T-pairs φ v 1 : v 1 → v 2 V 2 non ] and ((Π 1 ) v 1  (M 1 ) v 1 ) ((Π 2 ) v 2  (M 2 ) v 2 ) [where V 1 non κ κ [orbi-]Aut-holomorphic T-pairs φ v 1 : ((X 1 ) v 1  (M 1 ) v 1 ) ((X 2 ) v 2  (M 2 ) v 2 )  v 1 → v 2 V 2 arc ]. Write Th  [where V 1 arc T (respectively, Th T ) for the category of global (respectively, panalocal) T-pairs and morphisms of global (respectively, panalocal) T-pairs. Thus, it follows immediately from the definitions that we ob- tain a natural “panalocalization functor”  Th  T Th T lying over the functor Th  Th  of (iv) which is essentially surjective.  Moreover, we have compatible natural functors Th  EA  , Th  T EA , as well as natural functors  Th  TF Th TM ;  Th  TM Th TLG ;  Th  TF Th TM ;  Th  TM Th TLG [cf. Definition 3.1, (iii); Definition 4.1, (iii)]. Remark 5.1.1. Note that the reference isomorphism ψ V of Definition 5.1, (iii), is uniquely determined by the conditions stated. Indeed, for nonarchimedean elements, non this follows by considering the stabilizers in Π of elements of V , together with the well-known fact that a nonarchimedean prime of F [cf. the notation of Definition 5.1, (ii)] is uniquely determined by any open subgroup of its decomposition group in G F [cf., e.g., [NSW], Corollary 12.1.3]; for archimedean elements, this follows arc by considering the topology induced on k NF (Π) by A X v via “κ v for v V . arc Moreover, for v V , the isomorphism ψ v of Definition 5.1, (iii), is uniquely determined by the condition of compatibility with δ ell,v ell , δ v . Indeed, by Corollary 2.3, (i) [cf. also [Mzk14], Lemma 1.3, (iii)], this follows from the well-known fact that any automorphism of a hyperbolic orbicurve that induces the identity outer automorphism of the profinite fundamental group of the orbicurve is itself the identity automorphism. Similar uniqueness statements [with similar proofs] hold for the morphisms φ V , φ v of Definition 5.1, (iii). Corollary 5.2. (First Properties of Galois-theaters and Pairs) Let T {TF, TM}. We shall apply a subscript “TM” to [global or local] arithmetic data of TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 119 “T-pairs” to denote the result of applying the natural functor whose codomain is the corresponding category of “TM-pairs” [i.e., the identity functor if T = TM cf. Proposition 3.2]; we shall also use the subscript “TLG” in a similar way. (i) Write An  [Th  ] for the category whose objects are data of the form V  (Π) =  V  (Π), v } v∈V(Π) non , {(X(Π, v), δ ell,v , κ ell,v )} v∈V(Π) arc ) def [cf. the notation of Definition 5.1, (i)] for Π Ob(EA  ) and whose morphisms are the morphisms induced by morphisms of EA  . Then we have natural functors EA  An  [Th  ] Th  EA  where the first arrow is the functor obtained by assigning Ob(EA  ) Π → V  (Π); the second arrow is the functor obtained by forgetting the way in which the global Galois-theater data V  (Π) arose from Π; the third arrow is the natu- ral functor of Definition 5.1, (vi); the composite EA  EA  of these arrows is naturally isomorphic to the identity functor all of which are equivalences of categories. (ii) Let κ (V  , M  , v } v∈V , {(Π v  M v )} v∈V non , {(X v  M v )} v∈V arc ) be a global T-pair [as in Definition 5.1, (v)]. Then there is a unique [hence, in par- ticular, there exists a functorial relative to Th  T algorithm for constructing the] isomorphism  (M TM ) μ  (Π) μ  Z Z [cf. Theorem 1.9, (b); Remark 1.10.1, (ii)] of Π-modules that is compatible relative to the restriction morphisms v } v∈V non with the isomorphisms non ((M v ) TM ) μ  v ), for v V , obtained by composing the isomorphisms of μ  Z Z Corollary 1.10, (c); Remark 3.2.1. (iii) In the notation of (ii), there exists a functorial [i.e., relative to Th  T ] algorithm for constructing the Kummer map   gp   1 1 M TM (M TM ) M TLG → lim (M TM )) lim (Π)) −→ H (J, μ  −→ H (J, μ  Z Z J J where “J” ranges over the open subgroups of Π. In particular, the reference isomorphisms ψ  , v } of Definition 5.1, (v), are uniquely determined by the conditions stated in Definition 5.1, (v); in a similar vein, the isomorphisms φ  , v } that appear in the definition of a “morphism φ of global T-pairs” in Definition 5.1, (v), are uniquely determined by φ V  . (iv) Write An  [Th  T ] for the category whose objects are data of the form  M  T (Π) = (V (Π),M T  (Π), v (Π)} v∈V(Π) , κ {(Π v  M T (Π, v))} v∈V(Π) non , {(X(Π, v)  M T (Π, v))} v∈V(Π) arc ) def 120 SHINICHI MOCHIZUKI [cf. the notation of Definition 5.1, (v)] for Π Ob(EA  ) and whose morphisms are the morphisms induced by morphisms of EA  . Then we have natural functors   EA  An  [Th  T ] Th T EA where the first arrow is the functor obtained by assigning Ob(EA  Π → T ) (Π); the second arrow is the functor obtained by forgetting the way in which M  T the global T-pair data M  T (Π) arose from Π; the third arrow is the natural functor of Definition 5.1, (vi); the composite EA  EA  of these arrows is naturally isomorphic to the identity functor all of which are equivalences of categories that are [1-]compatible [in the evident sense] with the functors of (i). (v) Write An  [Th  ] for the category whose objects are data of the form V  (Π) = (Π, {V  (Π)}  ) def where V  (Π) is as in (i); we use the notation “{−}  to denote the data obtained by applying the panalocalization functor Th  Th  of Definition 5.1, (iv) for Π Ob(EA  ) and whose morphisms are the morphisms induced by morphisms of EA  . Then we have natural functors EA  An  [Th  ] Th  where the first arrow is the functor obtained by assigning Ob(EA  ) Π →  V (Π); the second arrow is the functor obtained by forgetting the way in which the panalocal Galois-theater data {V  (Π)}  arose from Π. Here, the first arrow EA  An  [Th  ] is an equivalence of categories. (vi) Write An  [Th  T ] for the category whose objects are data of the form   M  T (Π) = (Π, {M T (Π)} ) def  where M  T (Π) is as in (iv); we use the notation “{−} to denote the data obtained by applying the panalocalization functor Th  Th  of Definition 5.1, (vi) for Π Ob(EA  ) and whose morphisms are the morphisms induced by morphisms of EA  . Then we have natural functors  EA  An  [Th  T ] Th T where the first arrow is the functor obtained by assigning Ob(EA  ) Π →  M T (Π); the second arrow is the functor obtained by forgetting the way in which  arose from Π. Here, the first arrow EA  the panalocal T-pair data {M  T (Π)} An  [Th  T ] is an equivalence of categories. (vii) By replacing, in the definition of the objects of Th  T [cf. Definition 5.1, (iv)], the data in Orb(TG) (respectively, Orb(EA)) labeled by a(n) nonar- chimedean (respectively, archimedean) valuation by [the result of applying (−) T to] the data that constitutes the corresponding object of Orb(Anab) [cf. Definition 3.1, (vi)] (respectively, Orb(LinHol) [cf. Definition 4.1, (v)]), we obtain a category An  [Th  T ] TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 121 i.e., whose morphisms are the morphisms induced by morphisms of Th  together with natural functors   Th  An  [Th  T ] Th T Th where the first arrow is the functor arising from the definition of An  [Th  T ]; the second arrow is the “forgetful functor” [cf. the “forgetful functors” of assertion (ii) of Corollaries 3.6, 4.5]; the third arrow is the natural functor [cf. Definition 5.1, (vi)]; the composite Th  Th  of these arrows is naturally isomorphic to the identity functor all of which are equivalences of categories. Proof. In light of Remark 5.1.1, assertion (i) is immediate from the definitions and the results of §1, §2 [cf., especially, Theorem 1.9; Corollaries 1.10, 2.8] quoted in these definitions. Assertion (ii) follows, for instance, by comparing the given global T-pair with the global T-pair data M  T (Π) of assertion (iv) via the reference isomorphisms that appear in Definition 5.1, (v). In light of assertion (ii), assertion (iii) is immediate from the definitions [cf. also Proposition 3.2, (ii), (iv), at the κ nonarchimedean v; “κ v ”, the Kummer structure of “(X v  M v )” at archimedean v]. In light of assertion (iii), assertion (iv) is immediate from the definitions and the results of §1, §2 [cf., especially, Theorem 1.9; Corollaries 1.10, 2.8] quoted in these definitions. In a similar vein, assertions (v), (vi), and (vii) are immediate from the definitions and the results quoted in these definitions [cf. also Proposition 3.2, (ii), κ (iv), at the nonarchimedean v; “κ v ”, the Kummer structure of “(X v  M v )” at archimedean v]. Remark 5.2.1. Note that neither of the composite functors EA  Th  , EA  Th  T of Corollary 5.2, (v), (vi) is an equivalence of categories! Put another way, there is no natural, functorial way to “glue together” the various local data of a panalocal Galois-theater/T-pair so as so obtain a “global profinite group” that determines an object of EA  . Remark 5.2.2. By applying the equivalence EA  Th  of Corollary 5.2, (i), one may obtain a factorization EA  Th  An  [Th  T ] of the functor EA  An  [Th  T ] of Corollary 5.2, (iv). Thus, we obtain equivalences     An  [Th  of categories Th An [Th  T ] Th ; the functor Th T ] may be  thought of as a “global analogue” of the panalocal functor Th An  [Th  T ] of Corollary 5.2, (vii). Remark 5.2.3. A similar result to Corollary 5.2, (ii) [hence also similar re- sults to Corollary 5.2, (iii), (iv)], may be obtained when T = TLG, by using the archimedean primes, which are “immune” to the {±1}-indeterminacy of Proposi- arc tion 3.3, (i). Indeed, in the notation of Definition 5.1, (iii), (v), if v V , then by combining “κ ell,v ell with the isomorphism “ψ v arising from the reference iso- morphism of the global Galois-theater under consideration yields an inclusion of 122 SHINICHI MOCHIZUKI fields k NF (Π) → A X(Π,v ell ) A X v . On the other hand, by applying Corollary 1.10, (c); Remark 1.10.3, (ii), at any of the nonarchimedean elements of V , it follows (Π) may be related to the roots of unity of k NF (Π), while the restriction that μ  Z morphism at v of the global T-pair under consideration, together with the Kummer (M  ) to the roots of unity of A X v . Thus, structure at v, allow one to relate μ  Z we obtain a functorial algorithm [albeit somewhat more complicated than the al- gorithm discussed in Corollary 5.2, (ii)] for constructing the natural isomorphism (M  ) μ  (Π). μ  Z Z Definition 5.3. Let T {TF, TM, TLG}. We shall apply a subscript “TLG” (respectively, “TCG”) to arithmetic data of “T-pairs” to denote the result of ap- plying the natural functor whose codomain is the corresponding category of “TLG- (respectively, TCG) pairs” [cf. Proposition 3.2; Corollary 5.2]. In the following, the symbols ,  are to be understood as shorthand for the terms “multiplicative” and “additive”, respectively. Let κ M  = (V  , M  , v } v∈V , {(Π v  M v )} v∈V non , {(X v  M v )} v∈V arc ) def be a global T-pair [where V  is as in Definition 5.1, (iii)]. Thus, M  is equipped with a natural Aut(Π)-action [cf. Corollary 5.2, (iv); Remark 5.2.3]. In the follow- ing, we shall use a superscript profinite group to denote the sub-object of invariants def with respect to that profinite group; if v V = V /Aut(Π), then we shall write M v for the arithmetic data of the [orbi-]MLF-Galois/Aut-holomorphic T-pair indexed by v of the panalocal T-pair determined by M  , and Π v for the [orbi-]decomposition group of v. (i) Suppose that T = TLG. Then a -line bundle L  on M  is defined to be a collection of data (L  [ ]; [v] L  [v] TV } v∈V ) where L  [ ] is an (M  ) Π -torsor equipped with an Out(Π)-action that is com- patible with the natural Out(Π)-action on (M  ) Π and, moreover, factors through the quotient Out(Π)  Im(Out(Π) Out(Π/Δ)); for each v V , τ [v] L  [v] TV def is a trivialization of the torsor L  [v] TV over (M v Π v ) TV = M v Π v /(M v Π v ) TCG deter- mined by the M v Π v -torsor L  [v] obtained from L  [ ] via ρ v , for v V lying over v such that any element of L  [ ] determines [by restriction] the element of L  [v] TV given by τ [v], for all but finitely many v V . [Here, we note that the “[topological] value group” (M v Π v ) TV is equipped with a natural ordering [which may be used to define its topology] and is = R if v V arc ; moreover the nat- = Z if v V non and ural ordering on (M v Π v ) TV determines a natural ordering on L  [v] TV .] A morphism of -line bundles on M   ζ : L  1 L 2 TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 123  is defined to be an Out(Π)-equivariant isomorphism ζ[ ] : L  1 [ ] L 2 [ ] between the respective (M  ) Π -torsors such that each v V induces an isomorphism ζ[v] TV :   L  1 [v] TV L 2 [v] TV that maps τ 1 [v] to an element of L 2 [v] TV that is τ 2 [v]. Write  Th  T [M ] for the category of -line bundles on M  and morphisms of -line bundles on M  .  If φ : M  1 M 2 is a morphism of global T-pairs, then there is a natural pull-    back functor φ : Th  T [M 2 ] Th T [M 1 ]. In particular, the various categories  Th  T [M ] together form a fibered category Th  Th  T T   over Th  T , whose fibers are the categories Th T [M ]. Finally, we observe that one may generalize these definitions to the case of arbitrary T {TF, TM, TLG} by applying the subscript “TLG”, where necessary. (ii) Suppose that T = TF. Write O M  for the ring of integers of the field M  . Then an -line bundle L on M  is defined to be a collection of data (L [ ]; {| | L  [v] } v∈V arc ) Π where L [ ] is a rank one projective O M  -module equipped with an Out(Π)- action that is compatible with the natural Out(Π)-action on (M  ) Π and, moreover, factors through the quotient Out(Π)  Im(Out(Π) Out(Π/Δ)); for each v V arc , | | L  [v] is a Hermitian metric on the M v -vector space L [v] obtained from L [ ] (M  ) Π via ρ v , for v V lying over v. [Here, we recall that M v is an [orbi-]complex archimedean field.] In this situation, we shall also write L [v] for the M v Π v -vector space obtained from L [ ] (M  ) Π via ρ v , for v V lying over v V non . In Π Π particular, the O M  -module O M  , equipped with its usual Hermitian metrics at arc elements of V , determines an -line bundle which we shall refer to as the trivial -line bundle. A morphism of -line bundles on M  ζ : L 1 L 2 Π is defined to be a nonzero Out(Π)-equivariant morphism of O M ] :  -modules ζ[ arc L 1 [ ] L 2 [ ] such that for each v V , the induced isomorphism ζ[v] : L 1 [v] L 2 [v] maps integral elements [i.e., elements of norm 1] with respect to | | L  [v] to integral elements with respect to | | L  [v] . Write 1 2  Th  T [M ] for the category of -line bundles on M  and morphisms of -line bundles on M  .  If φ : M  1 M 2 is a morphism of global T-pairs, then there is a natural pull-    back functor φ : Th  T [M 2 ] Th T [M 1 ]. In particular, the various categories  Th  T [M ] “glue together” to form a fibered category Th  Th  T T 124 SHINICHI MOCHIZUKI   over Th  T , whose fibers are the categories Th T [M ]. Finally, the assignment [in the notation of the above discussion]    L [ ] → the (M TLG ) Π -torsor of nonzero sections of L [ ] (M  ) Π determines [in an evident fashion] an equivalence of categories Th  T Th  T over Th  T , i.e., an “equivalence of - and -line bundles”. (iii) Let  {, }; if  = , then assume that T = TF. Then observe  that the automorphism group of any object of Th  T [M ] is naturally isomorphic  ) Aut(Π) . To avoid various problems arising to the finite abelian group μ Q/Z (M TLG from these automorphisms, it is often useful to work with “coarsified versions” of the categories introduced in (i), (ii), as follows. Write | | Th T [M  ] for the [small, id-rigid!] category whose objects are isomorphism classes of objects   Aut(Π) of Th  -orbits of morphisms of T [M ] and whose morphisms are μ Q/Z (M TLG )    Th T [M ]. Thus, by allowing “M to vary, we obtain a fibered category | | Th T Th  T | | over Th  [M  ]. Finally, the equivalence of T , whose fibers are the categories Th T | | || categories of (ii) determines an equivalence of categories Th T Th T . Remark 5.3.1. In the notation of Definition 5.3, (iii), one may define in the style of Corollary 5.2, (iv) a category An  [Th  T , ||] whose objects are data of the form def | | | | [M  M T (Π) = (M  T (Π), Th T T [Π]]) for Π Ob(EA  ) and whose morphisms are the morphisms induced by morphisms | | of EA  . Here, we think of the datum “Th T [M  T [Π]]” as an object of the category whose objects are small categories with trivial automorphism groups and whose morphisms are contravariant functors. Then, just as in Corollary 5.2, (iv), one obtains a sequence of natural functors     EA  An  [Th  T , ||] An [Th T ] Th T EA where the first arrow is the functor obtained by assigning Ob(EA  T ) | | M T (Π) all of which are equivalences of categories. Definition 5.4. Π → Let T {TF, TM}, { , }. (i) If Z is an elliptically admissible hyperbolic orbicurve over an algebraic clo- sure of Q, then we shall refer to a hyperbolic orbicurve X as in Definition 5.1, (ii), TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 125 as geometrically isomorphic to Z if [in the notation of loc. cit.] there exists an isomorphism of schemes X F = Z. Write EA  [Z] EA  for the full subcategory determined by the profinite groups isomorphic to Π X for some X as in Definition 5.1, (i), that is geometrically isomorphic to Z. This full subcategory determines, in an evident fashion, full subcategories Th [Z] Th ; Th T [Z] Th T as well as full subcategories of the “An  [−]” versions of these categories discussed in Corollary 5.2 and the “-, -line bundle versions” discussed in Remark 5.3.1 [cf. also the “measure-theoretic versions” discussed in Remark 5.9.1 below]. (ii) By applying the functors “log T,T of Proposition 3.2, (v); Proposition 4.2, (ii), to the various local data of a panalocal T-pair, we obtain a panalocal log- Frobenius functor   log  T,T : Th T Th T which is naturally isomorphic to the identity functor, hence, in particular, an equiv- alence of categories. Note that the construction underlying this functor leaves the  underlying panalocal Galois-theater unchanged, i.e., log  T,T “lies over” Th . Now suppose that κ M  = (V  , M  , v } v∈V , {(Π v  M v )} v∈V non , {(X v  M v )} v∈V arc ) def is a global T-pair [where V  is as in Definition 5.1, (iii)]. Note that the various restriction morphisms ρ v determine a Π-equivariant embedding   M  → M v T v∈V of M  into a certain product of local data. Thus, by applying the functors “log T,T of Proposition 3.2, (v); Proposition 4.2, (ii), to the various local data of M  [i.e., more precisely: the data, other than the v }, that is indexed by v V ], we obtain a “local log-Frobenius functor log v T,T on the portion of a global T-pair constituted by this local data which is naturally isomorphic to the identity functor. More- over, by composing this natural isomorphism to the identity functor with the above embedding of M  , we obtain a new Π-equivariant embedding   M  → log v T,T (M v T ) v∈V  of M  into the product [as above] that arises from the output “log v T,T (M v T )” of log v T,T . In particular, by taking the image of this new embedding to be the global data Ob(T  ) [i.e., the “M  ”] of a new global T-pair whose local data is given by applying log v T,T to the local data of M  , we obtain a global log-Frobenius functor   log  T,T : Th T Th T 126 SHINICHI MOCHIZUKI which is naturally isomorphic to the identity functor, hence, in particular, an equiv- alence of categories. Moreover, the construction underlying this functor leaves the  underlying global Galois-theater unchanged, i.e., log  T,T “lies over” Th . In the fol- lowing discussion, we shall often denote [by abuse of notation] the restriction of log T,T to the categories “(−)[Z]” by log T,T . Note that if one restricts to the cat- egories “(−)[Z]”, then the set “V /Aut(Π)” has a meaning which is independent of the choice of a particular object of one of these categories [cf. the discussion of def Definition 5.1, (ii)]. In the following, let us fix a v V = V /Aut(Π). (iii) Consider, in the notation of Definition 3.1, (iv), the commutative diagram of natural maps O × → k  shell post-log... k id −→ k → × k  → k ...space-link × (k ) pf where we recall that k = (O × ) pf a diagram which determines an oriented k graph  Γ log [i.e., whose vertices and oriented edges correspond, respectively, to the non   objects and arrows of the above diagram]; write  Γ  non (respectively, Γ non ) for the oriented subgraph of  Γ log non obtained by removing the upper right-hand arrow “→ k” id (respectively, the lower left-hand arrow “k −→”) and  Γ × non for the intersection of   Γ    log non , Γ non . Let us refer to the lower left-hand vertex of Γ non [i.e., the first “k ”] log as the post-log vertex and to the other vertices of  Γ non as pre-log vertices; also we shall refer to the upper right-hand vertex of  Γ log non [i.e., “k”] as the space-link vertex. Here, we wish to think of the pre-log copy of “k as an object [i.e., “(O × ) pf ”] def k × formed from k and of the post-log copy of “k as the “new field” i.e., the new copy of the space-link vertex “k” obtained by applying the log-Frobenius functor. Observe that the entire diagram  Γ log non may be considered as a diagram in  the category TS, whereas the diagram  Γ non may be considered either as a diagram in the category TS or as a diagram in the category TS [i.e., relative to the additive topological group structure of the field k ]. Write p k for the residue characteristic def def of k; set p k = p k if p k is odd and p k = p 2 k if p k = 2. Then since [as is well-known] the p k -adic logarithm determines a bijection 1 + p k · O k p k · O k , it follows that O k Π k I = (p k ) −1 · I def (k ) Π k where the superscript “Π k denotes the submodule of Galois-invariants, and we write I for the image of O k × = (O × ) Π k O × via the left-hand vertical arrow k k of the above diagram, i.e., in essence, the compact submodule constituted by the pre-log-shell discussed in Definition 3.1, (iv). We shall refer to I as the log-shell of  Γ × non and to the left-hand vertical arrow of the above diagram as the shell-arrow. In fact, if k is absolutely unramified and p k is odd, then we have an equality O k Π k = I [cf. Remark 5.4.2 below]. MLF-sB MLF-sB (iv) Next, let us suppose that v V non ; recall the categories C TS , C TS MLF-sB MLF-sB of Definition 3.1, (iii). Thus, we have natural functors C TS C TS TG. TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 127 Let us write MLF-sB ) × Orb(TG),v Th [Z]; N v = Orb(C TS def MLF-sB N v = Orb(C TS ) × Orb(TG),v Th [Z] def where the “, v” in the fibered product is to be understood as referring to the natural functor Th [Z] Orb(TG) given by the assignment “V → Π v [cf. Defi- nition 5.1, (iv), (b)]. Thus, we have natural functors N v N v Th [Z]. Next, def  log  × def  × def     def   = Γ non , Γ v = Γ non ,  Γ  in the notation of (iii), let us set  Γ log v v = Γ non , Γ v = Γ non .  Then for each vertex ν of  Γ × v , by assigning to “k” or “O [i.e., depending on k the choice of T {TF, TM}] the object at the vertex ν of the diagram of (iii), we MLF-sB , hence by considering the portion of obtain a natural functor C T MLF-sB C TS the panalocal or global T-pair under consideration that is indexed by v or v V lying over v, a natural functor λ v,ν : Th T [Z] N v . In a similar vein, if ν is either  the space-link or the post-log vertex of  Γ log v , then by assigning to “k” or “O k the underlying additive topological group of the field “k” [cf. the functorial algorithms of Corollary 1.10, as applied in Proposition 3.2, (iii)], we obtain a natural functor MLF-sB , hence by considering the portion of the panalocal or global C T MLF-sB C TS T-pair under consideration that is indexed by v or v V lying over v, a natural functor λ v,ν : Th T [Z] N v . Thus, in summary, we obtain natural functors λ v,ν : Th T [Z] N v ; λ v,ν : Th T [Z] N v where the latter functor is obtained by composing the former functor with the natural functor N v N v that “lie over” Th [Z], for each vertex ν of  Γ log v . (v) Consider, in the notation of Definition 4.1, (iv), the commutative diagram of natural maps post-log... k id −→ k shell −→ k × → k ...space-link a diagram which determines an oriented graph  Γ log arc [i.e., whose vertices and oriented edges correspond, respectively, to the objects and arrows of the above    log diagram]; write  Γ  arc (respectively, Γ arc ) for the oriented subgraph of Γ arc obtained id by removing the arrow “→ k” on the right (respectively, the arrow “k −→” on     the left) and  Γ × arc for the intersection of Γ arc , Γ arc . Let us refer to the vertex of  log  Γ log arc given by the first “k as the post-log vertex and to the other vertices of Γ arc as pre-log vertices; also we shall refer to the vertex of  Γ log arc given by “k” as the space-link vertex. Here, we wish to think of the pre-log copy of “k as an object formed from k × and of the post-log copy of “k as the “new field” i.e., the new copy of the space-link vertex “k” obtained by applying the log-Frobenius functor. Observe that the entire diagram  Γ log arc may be considered as a diagram in   the category TH, whereas the diagram Γ arc may be considered either as a diagram in the category TH or as a diagram in TH [i.e., relative to the additive topological group structure of the field k ]. Note that it follows from well-known properties of the [complex] logarithm that O k = 1 ·I π I = O k × · I def k 128 SHINICHI MOCHIZUKI where we we write I for the uniquely determined “line segment” [i.e., more precisely: closure of a connected pre-compact open subset of a one-parameter sub- group] of k which is preserved by multiplication by ±1 and whose endpoints differ by a generator of Ker(k  k × ). Thus, I maps bijectively, except for the end- points of the line segment, to the pre-log-shell discussed in Definition 4.1, (iv). We × shall refer to I as the log-shell of  Γ × arc and to the arrow k  k as the shell-arrow. Also, we observe that I may be constructed as the closure of the union of the im- ages of I via the finite order automorphisms of the Aut-holomorphic group k ; in particular, the formation of I from I depends only on the structure of k as an object of TH. hol hol (vi) Next, let us suppose that v V arc ; recall the categories C TH , C TH of hol hol Definition 4.1, (iii). Thus, we have natural functors C TH C TH EA. Let us write hol ) × Orb(EA),v Th [Z]; N v = Orb(C TH def hol N v = Orb(C TH ) × Orb(EA),v Th [Z] def where the “, v” in the fibered product is to be understood as referring to the nat- ural functor Th [Z] Orb(EA) given by the assignment “V → X v [cf. Definition 5.1, (iv), (c)]. Thus, we have natural functors N v N v Th [Z]. Next, in the def  log  × def  ×   def     def   = Γ arc , Γ v = Γ arc , Γ v = Γ arc , Γ v = Γ arc . Then for notation of (v), let us set  Γ log v ×  each vertex ν of Γ v , by assigning to “k” or “O k  [i.e., depending on the choice of T {TF, TM}] the object at the vertex ν of the diagram of (v), we obtain a natural hol , hence by considering the portion of the panalocal or global functor C T hol C TH or T-pair under consideration that is indexed by v or v V lying over v, a natural functor λ v,ν : Th T [Z] N v . In a similar vein, if ν is either the space-link or the  post-log vertex of  Γ log v , then by assigning to “k” or “O k the underlying additive topological group of the field “k” [cf. the functorial algorithms of Corollary 2.7, as hol , hence by applied in Proposition 4.2, (ii)], we obtain a natural functor C T hol C TH considering the portion of the panalocal or global T-pair under consideration that is indexed by v or v V lying over v, a natural functor λ v,ν : Th T [Z] N v . Thus, in summary, we obtain natural functors λ v,ν : Th T [Z] N v ; λ v,ν : Th T [Z] N v where the latter functor is obtained by composing the former functor with the natural functor N v N v that “lie over” Th [Z], for each vertex ν of  Γ log v . (vii) Finally, in the notation of (iv) (respectively, (vi)) for v V non (respec-  log tively, v V arc ): For each edge  of  Γ  v (respectively, Γ v ) running from a vertex ν 1 to a vertex ν 2 , the arrow in the diagram of (iii) (respectively, (v)) corresponding to  determines a natural transformation ι v, : λ v,ν 1 Λ ν 1 λ v,ν 2 (respectively, ι v, : λ v,ν 1 Λ ν 1 λ v,ν 2 ) where, for each pre-log vertex ν of  Γ log v , we take Λ ν to be the identity functor on log  Th T [Z]; for the post-log vertex ν of Γ v , we take Λ ν to be the log-Frobenius functor log T,T : Th T [Z] Th T [Z]. TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 129 Remark 5.4.1. Note that the diagrams of Definition 5.4, (iii), (v), [hence also the natural transformations of Definition 5.4, (vii)] cannot be extended to global number fields! Indeed, this observation is, in essence, a reflection of the fact that the various logarithms that may be defined at the various completions of a number field do not induce maps from, say, the group of units of the number field to the number field! Remark 5.4.2. Note that in the context of Definition 5.4, (iii), when k is not absolutely unramified, the “gap” between O k Π k and I may be bounded in terms of the ramification index of k over Q p k . We leave the routine details to the interested reader. Remark 5.4.3. The inclusions “O k Π k I”, “O k I” of Definition 5.4, (iii), (v), may be thought of as inclusions, within the log-shell I, of the various localizations of the trivial -line bundle of Definition 5.3, (ii) an -line bundle whose structure is determined by the global ring of integers [i.e., “O M  in the notation of Definition 5.3, (ii)], equipped its natural metrics at the archimedean primes. That is to say, the definition of the trivial -line bundle involves, in an essential way, not just the additive [i.e., “”] structure of the global ring of integers, but also the multiplicative [i.e., “”] structure of the global ring of integers. Next, we consider the following global/panalocal analogue of Corollaries 3.6, 4.5. Corollary 5.5. (Global and Panalocal Mono-anabelian Log-Frobenius Compatibility) Let Z be an elliptically admissible hyperbolic orbicurve over an algebraic closure of Q, with field of moduli F mod [cf. Definition 5.1, (ii)]; T {TF, TM}; { , }. Set X = Th T [Z], E = Th [Z]. Consider the diagram of categories D def ... ... ... log X −→ id +1  λ  v   ...  X id   X  . . .  λ  v def log −→ X ...  id −1 ...   . . .  λ v  ... N v  N v   ... ... N v   ... N v  N v  ... ...  N v   ... ... E κ  An An [X ] ... 130 SHINICHI MOCHIZUKI  E where we use the notation “log” for the evident restriction of the arrows “log T,T of Definition 5.4, (ii); for positive integers n 7, we shall denote by D ≤n the subdiagram of categories of D determined by the first n [of the seven] rows of D ; we write L for the countably ordered set determined [cf. §0] by the infinite linear oriented graph  Γ opp D [so the elements of L correspond to vertices of the ≤1 first row of D ] and L = L {} def for the ordered set obtained by appending to L a formal symbol  [which we think of as corresponding to the unique vertex of the second row of D ] such that  < , for all  L; id  denotes the identity functor at the vertex  L; the vertices of the third and fourth rows of D are indexed by the elements v  , v, v  , . . . of the set of valuations V(F mod ) of F mod ; the arrows from the second row to the def category N v in the third row are given by the collection of functors λ v = v,ν } ν of Definition 5.4, (iv), (vi), where ν ranges over the pre-log vertices of  Γ log [or, v log  alternatively, over all the vertices of Γ v , subject to the proviso that we identify the functors associated to the space-link and post-log vertices]; the arrows from the third to fourth and from the fourth to fifth rows are the natural functors N v N v E of Definition 5.4, (iv), (vi); the arrows from the fifth to sixth and from the sixth to seventh rows are the natural equivalences of categories E An [X ] E , the first of which we shall denote by κ An , of Corollary 5.2, (i), (iv), (vii) [cf. also Remark 5.2.2], restricted to “[Z]”; we shall apply “[−]” to the names of arrows appearing in D to denote the path of length 1 associated to the arrow. Also, let us write φ An : An [X ] X for the equivalence of categories given by the “forgetful functor” of Corollary 5.2, (iv), (vii), restricted to “[Z]”, π An : X An [X ] for the quasi-inverse for φ An given by the composite of the natural projection functor X E with κ An : E An [X ], and η An : φ An π An id X for the isomorphism that exhibits φ An , π An as quasi-inverses to one another. Then: admits a natural structure of core on D ≤n−1 . That is (i) For n = 5, 6, 7, D ≤n to say, loosely speaking, E , An [X ] “form cores” of the functors in D. If, moreover, = , then one obtains a natural structure of core on D by appending to the final row of D the natural arrow E EA  [Z]. (ii) The “forgetful functor” φ An gives rise to a telecore structure T An on D ≤5 , whose underlying diagram of categories we denote by D An , by appending to D ≤6 telecore edges ... φ +1 ... X An [X ] φ    log −→ X log −→  φ −1 ... X ... TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III An [X ] φ  −→ 131 X from the core An [X ] to the various copies of X in D ≤2 given by copies of φ An , which we denote by φ  , for  L . That is to say, loosely speaking, φ An de- 0 . Finally, for each  L , let us write  ] termines a telecore structure on D ≤5 1 for the path on  Γ D An of length 0 at  and  ] for some [cf. the coricity of (i)!] path on  Γ D An of length {5, 6} [i.e., depending on whether or not  = ] that starts from , descends via some path of length {4, 5} to the core vertex “An [X ]”, and returns to  via the telecore edge φ  . Then the collection of natural transformations −1 −1 , η  , η  } ∈L,∈L  , η  where we write η  for the identity natural transformation from the arrow φ  : An [X ] X to the composite arrow id  φ  : An [X ] X and η  : (D An )  1 ] (D An )  0 ] for the isomorphism arising from η An generate a contact structure H An on the telecore T An . (iii) The natural transformations [cf. Definition 5.4, (vii)] ι v, : λ v,ν 1 Λ ν 1 λ v,ν 2 (respectively, ι v, : λ v,ν 1 Λ ν 1 λ v,ν 2 )  log where v V(F mod );  is an edge of  Γ  v (respectively, Γ v ) running from a vertex ν 1 to a vertex ν 2 ; if ν 1 is a pre-log vertex, then we interpret the domain and codomain of ι v, (respectively, ι v, ) as the arrows associated to the paths of length 1 (respectively, 2) from the second to third (respectively, fourth) rows of D determined by v and ν 1 , ν 2 ; if ν 1 is a post-log vertex, then we interpret the domain of ι v, (respectively, ι v, ) as the arrow associated to the path of length 3 (respectively, 4) from the first to the third (respectively, fourth) rows of D determined by v, ν 1 , and the condition that the initial length 2 portion of the path be a path of the form [id  ] [log] [for  L], and we interpret the codomain of ι v, (respectively, ι v, ) as the arrow associated to the path of length 2 (respectively, 3) from the first to the third (respectively, fourth) rows of D determined by v, ν 2 , and the condition that the initial length 1 portion of the path be a path of the form [id +1 ] [for the (respectively, D ≤4 ) same  L] belong to a family of homotopies on D ≤3 that determines on the portion of D ≤3 (respectively, D ≤4 ) indexed by v a structure of observable S log (respectively, S log ) on D ≤2 (respectively, the portion of D ≤3 indexed by v). Moreover, the families of homotopies that constitute S log and S log are compatible with one another as well as with the families of homotopies that constitute the core and telecore structures of (i), (ii). does not admit a structure of core on D ≤1 (iv) The diagram of categories D ≤2 which [i.e., whose constituent family of homotopies] is compatible with [the con- stituent family of homotopies of ] the observables S log , S log of (iii). Moreover, the telecore structure T An of (ii), the contact structure H An of (ii), and the observables S log , S log of (iii) are not simultaneously compatible. (v) The unique vertex  of the second row of D is a nexus of  Γ •D . More- over, D is totally -rigid, and the natural action of Z on the infinite linear 132 SHINICHI MOCHIZUKI oriented graph  Γ D ≤1 extends to an action of Z on D by nexus-classes of self- equivalences of D . Finally, the self-equivalences in these nexus-classes are com- patible with the families of homotopies that constitute the cores and observ- ables of (i), (iii); these self-equivalences also extend naturally [cf. the technique of extension applied in Definition 3.5, (vi)] to the diagram of categories [cf. Definition 3.5, (iv), (a)] that constitutes the telecore of (ii), in a fashion that is compatible with both the family of homotopies that constitutes this telecore structure [cf. Definition 3.5, (iv), (b)] and the contact structure H An of (ii). (vi) There is a natural panalocalization morphism of diagrams of categories D  D  [cf. the panalocalization functors of Definition 5.1, (iv), (vi)] that lies over the   evident isomorphism of oriented graphs  Γ  D Γ D and is compatible with the cores of (i), the telecore and contact structures of (ii), the observables of (iii), and the Z-actions of (v). Proof. Assertions (i), (ii) are immediate from the definitions cf. also the proofs of Corollary 3.6, (i), (ii); Corollary 4.5, (i), (ii). Next, we consider assertion (iii). def The data arising from applying the collection of functors λ v = v,ν } ν to the data arising from id  , as  ranges over the elements of L, yields a diagram of copies of  Γ log indexed by elements of L v ...  (  Γ log v ) +1  (  Γ log v )   (  Γ log v ) −1  ...  log where the symbol “” denotes the result of gluing (  Γ log v ) +1 onto ( Γ v )  by  log identifying the post-log vertex of (  Γ log v ) +1 with the space-link vertex of ( Γ v )  . Now the existence of a family of homotopies as asserted follows, in a routine fashion, from the fact that the above diagram is commutative [i.e., one does not obtain any pairs of distinct maps by traveling along distinct co-verticial pairs of paths of the diagram] cf. the relationship of the diagrams × ... ← k +1 ... ← × k +1 × (k +1 ) pf k +1 ← × k  × ← k  × (k  ) pf ← . . . k  ← . . . of Remarks 3.6.1, (i); 4.5.1, (i), to the proofs of assertion (iii) of Corollaries 3.6, 4.5; we leave the routine details [which are entirely similar to the proofs of assertion (iii) of Corollaries 3.6, 4.5] to the reader. Finally, the compatibility of the resulting family of homotopies with the families of homotopies that constitute the core and telecore structures of (i), (ii) is immediate from the definitions. This completes the proof of assertion (iii). Next, we consider assertion (iv). Recall that the proofs of the incompatibil- ity assertions of assertion (iv) of Corollaries 3.6, 4.5 amount, in essence, to the incompatibility of the introduction of a single model that maps isomorphically to various copies of the model indexed by elements of L in the diagrams of Remarks 3.6.1, (i); 4.5.1, (i). In the present situation, the incompatibility assertions of as- sertion (iv) of the present Corollaries 5.5 amount, in an entirely similar fashion, to TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 133  log that maps the incompatibility of the introduction of a single model (  Γ log v )  of Γ v isomorphically (  Γ log v )   ... ...  (  Γ log v ) +1  (  Γ log v )    (  Γ log v ) −1 ...  ... to each copy (  Γ log v )  that appears in the diagram that was used in the proof of assertion (iii). We leave the routine details [which are entirely similar to the proofs of assertion (iv) of Corollaries 3.6, 4.5] to the reader. This completes the proof of assertion (iv). Next, we consider assertion (v). The fact that  is a nexus of  Γ •D is immediate from the definitions. When = , the total -rigidity of D follows immediately  from the equivalence of categories Th  T EA of Corollary 5.2, (iv), together with the slimness of the profinite groups that appear as objects of EA  [cf., e.g., [Mzk20], Proposition 2.3, (ii)]. When = , the total -rigidity of D follows, in light of the “Kummer theory” of Propositions 3.2, (iv); 4.2, (i), from the fact that the orbi-objects that appear in the definition of a panalocal Galois-theater are defined in such a way as to eliminate all the automorphisms [cf. Definition 5.1, (iv), (b), (c)]. The remainder of assertion (v) is immediate from the definitions and constructions made thus far. This completes the proof of assertion (v). Finally, we observe that assertion (vi) is immediate from the definitions and constructions made thus far. Remark 5.5.1. The “general formal content” of the remarks following Corollaries 3.6, 3.7 applies to the situation discussed in Corollary 5.5, as well. We leave the routine details of translating these remarks into the language of the situation of Corollary 5.5 to the interested reader. Remark 5.5.2. Note that it does not appear realistic to attempt to construct a theory of “geometric panalocalization” with respect to the various closed points of the hyperbolic orbicurve over an MLF under consideration [cf. the discussion of Remarks 1.11.5; 3.7.7, (ii)]. Indeed, the decomposition groups of such closed points [which are either isomorphic to the absolute Galois group of an MLF or an extension  of such an absolute Galois group by a copy of Z(1)] do not satisfy an appropriate analogue of the crucial mono-anabelian result Corollary 1.10 [hence, in particular, do not lead to a situation in which both of the two combinatorial dimensions of the absolute Galois group of an MLF under consideration are rigidified cf. Remark 1.9.4]. Definition 5.6. (i) Recall the categories TG, TM, and TS of Definition 3.1, (i), (iii). Write TG TG 134 SHINICHI MOCHIZUKI for the subcategory given by the profinite groups isomorphic to the absolute Galois group of an MLF and open injections of profinite groups, and TG cs Orb(TG) for the full subcategory determined by the [“coarsified”] objects of Orb(TG) obtained by considering an object G Ob(TG) up to its group of automorphisms Aut TG (G). Write TM for the category whose objects (C, C ) consist of a topological monoid C isomorphic to O C  and a topological submonoid C C [necessarily isomorphic to R ≥0 ] such that the natural inclusions C × → C [where C × , which is necessarily isomorphic to S 1 , denotes the topological submonoid of invertible elements], C → C determine an isomorphism C × × C C of topological monoids, and whose morphisms (C 1 , C 1 ) (C 2 , C 2 ) are isomor- phisms of topological monoids C 1 C 2 that induce isomorphisms C 1 C 2 . If G Ob(TG), then let us write Lie(G) for the associated group germ i.e., the associated group pro-object of TS de- termined by the neighborhoods of the identity element and, when G is abelian, Lie ± (G) for the orbi-group germ obtained by working with Lie(G) up to “{±1}”. Write TB for the category whose objects (B, B  , B  , β) consist of a two-dimensional connected topological Lie group B equipped with two one-parameter subgroups B  , B  B that determine an isomorphism B  × B  B of topological groups, together with an isomorphism β : Lie ± (B  ) Lie ± (B  ), and whose morphisms (B 1 , B 1  , B 1  , β 1 ) (B 2 , B 2  , B 2  , β 2 ) are the surjective homomor- phisms B 1 B 2 of topological groups that are compatible with the B i  , B i  , β i for i = 1, 2. Write TB for the category of orientable topological orbisurfaces [i.e., which are topological surfaces over the complement, in the “coarse space” associated to the orbisurface, of some discrete closed subset] and local isomorphisms between such orbisurfaces. Thus, we obtain natural “forgetful functors” TM TM; TB TB determined by the assignments (C, C ) → C, (B, B  , B  , β) → B, as well as natural “decomposition functors” dec TM  : TM TG × TG cs ; dec TB : TB TG × TG cs gp determined by the assignments (C, C ) → (C × , ( C ) cs ), (B, B  , B  , β) → (B  , (B  ) cs ) [where “gp” denotes the groupification of a monoid; “cs” denotes the result of con- sidering a topological group up to its group of automorphisms]. TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 135 (ii) We shall refer to as a mono-analytic Galois-theater any collection of data W = (W  , {(G w , |Π| w )} w∈W non , {(G w , |X| w )} w∈W arc ) def where W  is a set that admits a decomposition as a disjoint union W  = def { W } W arc W = W non W arc ; for each w W non , G w W non Ob(Orb(TG )), and |Π| w is an isomorphism class of pro-objects of the category TG; for each w W arc , G w Ob(Orb(TM )), and |X| w is an isomorphism class of EA such that there exists a panalocal Galois-theater V  = (V  , v } v∈V non , {X v } v∈V arc ) def [cf. the notation of Definition 5.1, (iv)] and an isomorphism of sets ψ W : V  W  which we shall refer to as a reference isomorphism for W that satisfies the following conditions: (a) ψ W maps V → W , V non W non , V arc W arc ; (b) for each v V non mapping to w W non , G w is isomorphic to the [group- theoretically characterizable cf. Remark 1.9.2] quotient Π v  G v determined by the absolute Galois group of the base field, and the class |Π| w contains the pro-object of TG determined by the projective system of open subgroups of Π v arising from open subgroups of G v ; (c) for each v V arc mapping to w W arc , X v belongs to the class |X| w , and G w is isomorphic to the object of TM determined   R >0 ). A morphism of mono-analytic Galois-theaters , O A by (O A X v X v φ : (W 1  ,{((G 1 ) w 1 , |Π| w 1 )} w 1 ∈W 1 non , {((G 1 ) w 1 , |X| w 1 )} w 1 ∈W 1 arc ) (W 2  , {((G 2 ) w 2 , |Π| w 2 )} w 2 ∈W 2 non , {((G 2 ) w 2 , |X| w 2 )} w 2 ∈W 2 arc ) is defined to consist of a bijection of sets φ W : W 1  W 2  that induces bijec- tions W 1 non W 2 non , W 1 arc W 2 arc that are compatible with the isomorphism classes |Π| w i , |X| w i [for i = 1, 2], together with open injections of [orbi-]profinite w 1 → w 2 W 2 non ], and isomorphisms groups (G 1 ) w 1 → (G 2 ) w 2 [where W 1 non (G 1 ) w 1 (G 2 ) w 2 [where W 1 arc w 1 → w 2 W 2 arc ]. Write Th for the category of mono-analytic Galois-theaters and morphisms of mono-analytic Galois-theaters. Thus, if Z is an elliptically admissible hyperbolic orbicurve over an algebraic clo- sure of Q, then we have a full subcategory Th [Z] Th , together with natural “mono-analyticization functors” Th  Th ; Th  [Z] Th [Z] which are essentially surjective. (iii) Next, let us fix a mono-analytic Galois-theater W as in (ii), together with MLF MLF , C TS of Definition 3.1, (iii). Thus, we a w W non . Recall the categories C TS have a [1-]commutative diagram of natural functors MLF-sB C TS  MLF-sB −→ C TS  −→ TG sB  MLF C TS −→ −→ MLF C TS TG 136 SHINICHI MOCHIZUKI in which the vertical arrows are “mono-analyticization functors” [cf. the mono- analyticization functors of (ii); the construction implicit in (ii), (b)]; the arrows MLF-sB MLF TG sB , C TS TG are the natural projection functors. Let us write C TS def MLF = Orb(C TS ) × Orb(TG  ),w Th [Z] N w def MLF N w = Orb(C TS ) × Orb(TG  ),w Th [Z] where the “, w” in the fibered product is to be understood as referring to the natural functor Th [Z] Orb(TG ) given by the assignment “W → G w [cf. (ii)]. Thus, we have natural functors N w N w Th [Z]. (iv) Next, let us fix a mono-analytic Galois-theater W as in (ii), together with hol hol , C TH of Definition 4.1, (iii). Write a w W arc . Recall the categories C TH hol C TB for the category whose objects are triples (G, M, κ M ), where G Ob(TM ), M Ob(TB), and κ M : dec TB (M )  dec TM  (G) which we shall refer to as the Kummer structure of the object is a pair of surjective homomorphisms of TG, TG cs , and whose morphisms φ : (G 1 , M 1 , κ M 1 ) (G 2 , M 2 , κ M 2 ) consist of an isomorphism φ G : G 1 G 2 of TM and a morphism φ M : M 1 M 2 of TB that hol def are compatible with κ M 1 , κ M 2 ; write C TB = TM × TB. Next: κ hol ) [cf. Definition 4.1, (i)]. Recall Suppose that (X ell  M k ) Ob(C TH κ that the Kummer structure of (X ell  M k ) consists of an Aut-holomorphic homomorphism from M k to an isomorph of “C × ( = S 1 × R >0 )”; observe that the Aut-holomorphic automorphisms of Lie(C × ) of order 4 determine an isomorphism Lie ± (S 1 ×{1}) Lie ± ({1}×R >0 ). Thus, by pulling back κ to M k , via the Kummer structure of (X ell  M k ), the two one-parameter subgroups “S 1 × {1}, {1} × R >0 C × ”, we obtain, in a natural way, an hol . object of C TB In particular, we obtain a [1-]commutative diagram of natural functors hol C TH  −→ hol C TB hol −→ C TB hol C TH  −→ EA  −→ TM in which the vertical arrows are “mono-analyticization functors” [cf. the mono- analyticization functors of (ii); the construction implicit in (ii), (c)]; the arrows hol hol EA, C TB TM are the natural projection functors. Let us write C TH N w def hol = Orb(C TB ) × Orb(TM  ),w Th [Z] def hol ) × Orb(TM  ),w Th [Z] N w = Orb(C TB where the “, w” in the fibered product is to be understood as referring to the natural functor Th [Z] Orb(TM ) given by the assignment “W → G w [cf. (ii)]. Thus, we have natural functors N w N w Th [Z]. TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 137 Remark 5.6.1. (i) Observe that a monoid M may be thought of as a [1-]category C M consisting of a single object whose monoid of endomorphisms is given by M . In a similar vein, a ring R, whose underlying additive group we denote by R , may be thought of as a 2-category consisting of the single 1-category C R  , together with the functors C R  C R  arising from left multiplication by elements of R. (ii) The constructions of (i) suggest that whereas a monoid may be thought of as a mathematical object with “one combinatorial dimension”, a ring should be thought of as a mathematical object with “two combinatorial dimensions”. More- over, in the case of an MLF k, these two combinatorial dimensions may be thought of as corresponding to the two cohomological dimensions of the absolute Galois group of k, while in the case of a CAF k, these two combinatorial dimensions may be thought of as corresponding to the two real or topological dimensions of k. Thus, from this point of view, it is natural to think of ring structures as corresponding to holomorphic structures i.e., both ring and holomorphic structures are based on a certain complicated “intertwining of the underlying two combinatorial dimensions”. So far, in the theory of §1, §2, §3, and §4 of the present paper, the emphasis has been on “holomorphic structures”, i.e., of restricting ourselves to sit- uations in which this “complicated intertwining” is rigid. By contrast, the various ideas introduced in Definition 5.6 relate to the issue of disabling this rigidity i.e., of “passing from one holomorphic to two underlying combinatorial/topological dimensions” an operation which, as was discussed in Remarks 1.9.4, 2.7.3, has the effect of leaving only one of the two combinatorial dimensions rigid. Put an- other way, this corresponds to the operation of “passing from rings to monoids”; this is the principal motivation for the term “mono-analyticization”. The following result is elementary and well-known. Proposition 5.7. (Local Volumes) Let k be either a mixed-characteristic nonarchimedean local field or a complex archimedean field. (i) Suppose that k is nonarchimedean [cf. Definition 3.1, (i)]. Write m k O k for the maximal ideal of O k and M(k) for the set of compact open subsets of k. Then: (a) There exists a unique map μ k : M(k) R >0 that satisfies the following properties: (1) additivity, i.e., μ k (A B) = μ k (A) + μ k (B), for A, B M(k) such that A B = ∅; (2) -translation invariance, i.e., μ k (A + x) = μ k (A), for A M(k), x k; (3) normal- ization, i.e., μ k (O k ) = 1. We shall refer to μ k (−) as the volume on k. def Also, we shall write μ log k (−) = log(μ k (−)) [where log denotes the natural logarithm R >0 R] and refer to μ log k (−) as the log-volume on k. If the residue field of k is of cardinality p f , where p is a prime number and f n a positive integer, then, for n Z, μ log k (m k ) = −f · n · log(p). 138 SHINICHI MOCHIZUKI def def (b) Let x k × ; set μ̇ k (x) = μ k (x · O k ), μ̇ log k (x) = log( μ̇ k (x)). Then for log log × A M(k), we have μ k (x·A) = μ k (A)+ μ̇ log k (x); in particular, if x O k , log then μ log k (x · A) = μ k (A). (c) Write log k : O k × k for the [p-adic] logarithm on k. Let A O k × be an open subset such that log k induces a bijection A log k (A). Then log μ log k (A) = μ k (log k (A)). (ii) Suppose that k is archimedean [cf. Definition 4.1, (i)]; thus, we have a natural decomposition k × = O k × × R >0 , where O k × = S 1 , and we note that the projection k × R >0 extends to a continuous map pr R : k R. Write M(k) (respectively, M̆(k)) for the set of nonempty compact subsets A k (respectively, A k × ) such that A projects to a [compact] subset of R (respectively, O k × ) which is the closure of its interior in R (respectively, O k × ). Then: (a) The standard R-valued absolute value on k determines a Riemannian metric [as well as a Kähler metric] on k that restricts to Riemannian met- rics on O k × O k × × {1} → k × and R >0 {1} × R >0 → k × . Integrating these metrics over the projection of A M(k) (respectively, A M̆(k)) to R (respectively, O k × ) [i.e., “computing the length of A relative to these metrics”] yields a map μ k : M(k) R >0 (respectively, μ̆ k : M̆(k) R >0 ) that satisfies the following properties: (1) additivity, i.e., μ k (A B) = μ k (A) + μ k (B) (respectively, μ̆ k (A B) = μ̆ k (A) + μ̆ k (B)), for A, B M(k) (respectively, A, B M̆(k)) whose projections to R (respectively, O k × ) are disjoint; (2) normalization, i.e., μ k (O k ) = 1 (respectively, μ̆ k (O k × ) = 2π). We shall refer to μ k (−) (respectively, μ̆ k (−)) as the ra- dial volume (respectively, angular volume) on k. Also, we shall write def def log μ log k (−) = log(μ k (−)) (respectively, μ̆ k (−) = log(μ̆ k (−))) and refer to log μ log k (−) (respectively, μ̆ k (−)) as the radial log-volume (respectively, angular log-volume) on k. def def (b) Let x k × ; set μ̇ k (x) = μ k (x · O k ), μ̇ log k (x) = log( μ̇ k (x)). Then for log log A M(k) (respectively, A M̆(k)), we have μ log k (x·A) = μ k (A)+ μ̇ k (x) log log × (respectively, μ̆ log k (x · A) = μ̆ k (A)); in particular, if x O k , then μ k (x · A) = μ log k (A). (c) Write exp k : k k × for the exponential map on k. Let A M(k) be such that exp k (A) O k × , and, moreover, the maps pr R and exp k induce log bijections A pr R (A), A exp k (A). Then μ log k (A) = μ̆ k (exp k (A)). TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 139 Proof. First, we consider assertion (i). Part (a) follows immediately from well- known properties of the Haar measure on the locally compact [additive] group k. Part (b) follows immediately from the uniqueness portion of part (a). To verify part (c) for arbitrary A, it suffices [by the additivity property of μ k (−)] to verify part (c) for A of the form x + m nk for n a sufficiently large positive integer, x O k × . But then log k determines a bijection x + m nk log k (x) + m nk , so the equality log log μ log k (A) = μ k (log k (A)) follows from the -translation invariance of μ k (−). This completes the proof of assertion (i). Assertion (ii) follows immediately from well- known properties of the geometry of the complex plane. Remark 5.7.1. The “log-compatibility” [i.e., part (c)] of Proposition 5.7, (i), (ii), may be regarded as a sort of “integrated version” of the fact that the derivative of the formal power series log(1 + X) = X + . . . at X = 0 is equal to 1. Moreover, the opposite directions of the “arrows involved” [i.e., logarithm versus exponential] in the nonarchimedean and archimedean cases is reminiscent of the discussion of Remark 4.5.2. Proposition 5.8. (Mono-analytic Reconstruction of Log-shells) (i) Let G k be the absolute Galois group of an MLF k. Then there exists a functorial [i.e., relative to TG ] “group-theoretic” algorithm for constructing × ab the images of the embeddings O k  → G ab k , k → G k of local class field theory [cf. [Mzk9], Proposition 1.2.1, (iii), (iv)]. Here, the asserted “functoriality” is con- travariant and induced by the Verlagerung, or transfer, map on abelianizations. In particular, we obtain a functorial “group-theoretic” algorithm for reconstructing the residue characteristic p [cf. [Mzk9], Proposition 1.2.1, (i)], the invariant p [i.e., p if p is odd; p 2 if p is even cf. Definition 5.4, (iii)], the cardinality p f of the residue field of k [i.e., by adding 1 to the cardinality of the prime-to-p torsion of k × ], the absolute degree [k : Q p ] [i.e., as the dimension of O k × Q p over Q p ], the absolute ramification index e = [k : Q p ]/f , and the order p m of the subgroup of p-th power roots of unity of k × . (ii) The algorithms of (i) yield a functorial [i.e., relative to TG ] “group- theoretic” algorithm “Ob(TG ) G →  Γ × non (G)” for constructing from G the MLF  Γ × [cf. §0] non -diagram in C TS × O × (G) k  → k (G)  k (G) → (k ) pf (G) × determined by the diagram of Definition 5.4, (iii), hence also the log-shell I(G) k (G) of  Γ × non (G). (iii) The algorithms of (i) yield a functorial [i.e., relative to TG ] “group- theoretic” algorithm “Ob(TG ) G → R non (G)” for constructing from G the topological group [which is isomorphic to R] × R non (G) = (k (G)/O × (G)) def k 140 SHINICHI MOCHIZUKI where “∧” stands for the completion with respect to the order structure deter- mined by the nonnegative elements, i.e., the image of O  (G)/O × (G) equipped k k with a distinguished element, namely, the “Frobenius element” F(G) R non (G) [cf. [Mzk9], Proposition 1.2.1, (iv)], which we think of as corresponding to the el- ement f G · log(p G ) R, where p G , f G are the invariants “p”, “f of (i). Finally, these algorithms also yield a functorial, “group-theoretic” algorithm for constructing the log-volume map μ log (G) : M(k (G) G ) R non (G) where “M(−)” is as in Proposition 5.7, (i); the superscript “G” denotes the submodule of G-invariants; if we write m G , e G , p G for the invariants “m”, “e”, “p of (i), then one may think of μ log (G) as being normalized via the formula μ log (G)(I(G)) = {−1 m G /f G + e G · log(p G )/log(p G )} · F(G) determined by composing the map μ log of Proposition 5.7, (i), (a), with the k (G) G isomorphism R R non (G) given by f G ·log(p G ) → F(G). That is to say, “μ log (G)” and “M(k (G) G )” are well-defined despite the fact one does not have an algorithm for reconstructing the field structure on k (G) G [i.e., unlike the situation discussed in Proposition 5.7, (i)]. (iv) Let G = (C, C ) Ob(TM ); write C C × for the [pointed] universal covering of C × [cf. the definition of “k  k × in Definition 4.1, (i)]; thus, we regard C as a topological group [isomorphic to R]. Then the evident isomorphism def def Lie ± (C ) = Lie ± (C × ) allows one to regard k (G) = C ×C , k × (G) = C × ×C as objects of TB. Write Seg(G) for the equivalence classes of compact line segments on C [i.e., compact subsets which are either equal to the closure of a connected open set or are of cardinality one], relative to the equivalence relation determined by translation on C . Then forming the union of two compact line segments whose intersection is of cardinality one determines a monoid structure on Seg(G) with respect to which Seg(G) R ≥0 . In particular, this monoid structure determines a structure of topological monoid on Seg(G). (v) The constructions of (iv) yield a functorial [i.e., relative to TM ] algo-  × rithm “Ob(TM ) G →  Γ × arc (G)” for constructing from G the Γ arc -diagram in hol [cf. §0] C TB k (G) = C × C  k × (G) = C × × C determined by the diagram of Definition 5.4, (v), hence also the log-shell I(G) = {(a · x, b · x) | x I C ; a, b R; a 2 + b 2 = 1} k (G) def where we write I C C for the unique compact line segment on C that is invariant with respect to the action of ±1 and, moreover, maps bijectively, except for its endpoints, to C × of  Γ × arc (G). TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 141 (vi) The constructions of (iv) yield a functorial [i.e., relative to TM ] al- G → R arc (G)” for constructing from G the topological gorithm “Ob(TM ) group [which is isomorphic to R] def R arc (G) = Seg(G) gp equipped with a distinguished element, namely, the “Frobenius element” F(G) Seg(G) R arc (G) determined by a compact line segment that maps bijectively, except for its endpoints, to C × ; we shall think of F(G) as corresponding to R. Finally, these algorithms also yield a functorial, algorithm for constructing the radial and angular log-volume maps μ log (G) : M(k (G)) R arc (G); μ̆ log (G) : M̆(k (G)) R arc (G) where “M(−)”, M̆(−)” are as in Proposition 5.7, (ii); if, in the style of the definition of I(G) in (v), we write ∂I C for the boundary [i.e., the two endpoints] of I C and O k × (G) = {(a · x, b · x) | x ∂I C ; a, b R; a 2 + b 2 = π −2 } k (G) def [so one has a natural bijection R >0 × O k × (G) k (G)\{0}], then one may think of μ log (G), μ̆ log (G) as being normalized via the formulas μ log (G)(I(G)) = μ̆ log (G)(O k × (G)) log(2) · F(G)/2π = log(π) · F(G)/2π log determined by composing the maps μ log k (G) , μ̆ k (G) of Proposition 5.7, (ii), (a), with the isomorphism R R arc (G) given by → F(G). That is to say, “μ log (G)”, “μ̆ log (G)”, “M(k (G))”, and M̆(k (G))” are well-defined despite the fact one does not have an algorithm for reconstructing the field structure on k (G) [i.e., unlike the situation discussed in Proposition 5.7, (ii)]. (vii) Let Z be an elliptically admissible hyperbolic orbicurve over an alge- braic closure of Q; V  Ob(Th  [Z]) [cf. the notation of Definition 5.1, (iv); Definition 5.4, (i)]; W Ob(Th [Z]) the mono-analyticization of V  [cf. the notation of Definition 5.6, (ii)]; w W non (respectively, w W arc ). Write An [N w ] for the category whose objects consist of an object of Th [Z], together with the MLF  × hol  × object of Orb(C TS [ Γ non ]) (respectively, Orb(C TB [ Γ arc ])) given by applying the × algorithm “G →  Γ non (G)” of (ii) (respectively, “G →  Γ × arc (G)” of (v)) to the object of Orb(TG ) (respectively, Orb(TM )) obtained by projecting [at w cf. Defini- tion 5.6, (ii)] the given object of Th [Z], and whose morphisms are the morphisms induced by Th [Z]. Thus we obtain a natural equivalence of categories Th [Z] An [N w ] together with a “forgetful functor”  An ψ w,ν : An [N w ] N w 142 SHINICHI MOCHIZUKI def  × [cf. Definition 5.6, (iii), (iv)] for each vertex ν of  Γ × w = Γ non (respectively, def  ×  Γ × w = Γ arc ), and a natural transformation  An ι w,   An An : ψ w,ν ψ w,ν 1 2 for each edge  of  Γ × w running from a vertex ν 1 to a vertex ν 2 . Finally, we shall omit the symbol “” from the above notation to denote the result of composing the functors and natural transformations discussed above with the natural functor N w N w ; also, we shall replace the symbol “An by the symbol “” in the superscripts of the above notation to denote the result of restricting the functors and natural transformations discussed above to Th [Z]. Proof. The various assertions of Proposition 5.8 are immediate from the definitions and the references quoted in these definitions. Remark 5.8.1. (i) One way to summarize the archimedean portion of Proposition 5.8 is as follows: Suppose that one starts with the [Aut-]holomorphic monoid given by an isomorph of O C  [i.e., where one thinks of the [Aut-]holomorphic structure on O C  as  consisting of a(n) [Aut-]holomorphic structure on (O C  ) gp = C × ] arising as the O A X for some X Ob(EA). The operation of mono-analyticization consists of “forgetting the rigidification of the [Aut-]holomorphic structure furnished by X” [cf. Remark 2.7.3]. Thus, applying the operation of mono-analyticization to an isomorph of O C  yields the object of TM consisting of an isomorph of the topological monoid O C  equipped with the submonoid corresponding to O C  R >0 , which is non-rigid, in the sense that it is subject to dilations [cf. Remark 2.7.3]. On the other hand: From the point of view of the theory of log-shells, one wishes to per- form the operation of mono-analyticization i.e., of “forgetting the [Aut- ]holomorphic structure” in such a way that one does not obliterate the metric rigidity [i.e., the “applicability” of the theory of Proposition 5.7] of the log-shells involved. This is precisely what is achieved by the use of the category TB cf., especially, hol hol C TB in Definition 5.6, (iv); the the construction of the natural functor C TH constructions of Proposition 5.8, (iv), (v), (vi). That is to say, the “metric rigidity” of log-shells is preserved even after mono-analyticization by thinking of the “metric rigidity” of the original [Aut-]holomorphic O C  as being constituted by “the metric rigidity of S 1 = O C × , together with the rotation automor- phisms of Lie(C × ) of order 4” [cf. Definition 5.6, (iv)]. That is to say, this approach to describing “[Aut-]holomorphic metric rigidity” has the advantange of being “immune to mono-analyticization” cf. the construction of k (G) as “C × C in Proposition 5.8, (iv). On the other hand, it has the TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 143 disadvantage that it is not compatible [as one might expect from any sort of mono- analyticization operation!] with preserving the complex archimedean field structure of “k ”. That is to say, the two factors of C appearing in the product “C × C which should correspond to the imaginary and real portions of such a complex archimedean field structure may only be related to one another up to a {±1} indeterminacy, an indeterminacy that has the effect of obliterating the ring/field structure involved. (ii) It is interesting to note that the discussion of the archimedean situation of (i) is strongly reminiscent of the nonarchimedean portion of Proposition 5.8, which allows one to construct metrically rigid log-shells which are immune to mono-analyticization, but only at the expense of sacrificing the ring/field structures involved. Definition 5.9. (i) By pulling back the various functorial algorithms of Proposition 5.8 defined on TG , TM via the mono-analyticization functors TG sB TG , EA TM , we obtain functorial algorithms defined on TG sB , EA. In particular, if, in the notation of Definition 5.1, (iii) (respectively, Definition 5.1, (iv); Definition 5.6, (ii)), V  (respectively, V  ; W ) is a global (respectively, panalocal; mono-analytic) Galois- def theater, then for each v V = V /Aut(Π) (respectively, v V ; v W ), we obtain i.e., by applying the functorial algorithms “R non (−)”, “R arc (−)” of Proposition 5.8, (iii), (vi) [orbi-]topological groups [isomorphic to R] R v equipped with distinguished Frobenius elements F v R v . Moreover, if we write V  def for the unique global element of V  = V /Aut(Π) (respectively, V  ; W  ), then we obtain a(n) [orbi-]topological group [isomorphic to R]  R  V R v v where the product ranges over v V (respectively, v V ; v W ) obtained as the “graph” of the correspondences between the R v ’s that relate the F v /(f v · log(p v )) [where “f v ”, “p v are the invariants “f G ”, “p G of Proposition 5.8, (iii)] for nonarchimedean v to the F v /2π for archimedean v. Thus, R  V is equipped with a distinguished element F  V R  V [which we think of as corresponding to 1 R], and we have natural isomorphisms of [orbi-]topological groups R  V R v that map F  V → F v /(f v · log(p v )) for nonarchimedean v and F  V → F v /2π for archimedean v [where we note that division of elements of the abstract topological group R v by a positive real number is well-defined]. (ii) In the notation of Definition 5.1, (v) (respectively, Definition 5.1, (vi)), let M  (respectively, M  ) be a global (respectively, panalocal) T-pair, for T def  {TF, TM}. In the non-resp’d case, write V  = V /Aut(Π). Then the various log-volumes defined in Proposition 5.7, (i), (ii), determine maps Π v log v : M(M v ) R v } v∈V ; Π v {μ̆ log v : M̆(M v ) R v } v∈V arc 144 SHINICHI MOCHIZUKI where we write M(M v Π v ), [when v V arc ] M̆(M v Π v = M v ) for the set of sub- sets determined [via the reference isomorphisms “ψ v of Definition 5.1, (v); the “forgetful functors” of Corollary 5.2, (iv), (vii)] by intersecting with M T (Π, v) Π v k NF (Π, v) Π v the corresponding collection of subsets of M(k NF (Π, v) Π v ), [when v × V arc ] M̆(k NF (Π, v) Π v ). (iii) In the non-resp’d [i.e., global] case of (ii), suppose further that T = TF. Then for any -line bundle L on M  , one verifies immediately that there exist morphisms of -line bundles on M  ζ : L 1 L ; ζ 0 : L 1 L 0 such that L 0 is isomorphic to the trivial -line bundle. Thus, for each v V , we obtain isomorphisms of M v Π v -vector spaces ζ[v] : L 1 [v] L [v], ζ 0 [v] : L 1 [v] L 0 [v]. Moreover, by applying these isomorphisms, we obtain subsets S v L 0 [v] for each v V as follows: If v V non , then we take S v to be the subset determined by the closure of the image [via the various ρ v , for v V lying over v] of L [ ]. If v V arc , then we take S v to be the subset determined by the set of elements of L [v] for which | | L  [v] 1. Now set   def  mod  mod μ log log + μ log R  V  (L ) = v (S v ) /d v v (S v ) /d v v∈V arc v∈V non where d mod is as in Definition 5.1, (ii), for v V = V(F mod ); the superscript v denotes the result of applying the natural isomorphisms R  V R v of (i); we note that the sum is finite, since μ log v (S v ) = 0 for all but finitely many v V . As is well-known [or easily verified!] from elementary number theory i.e., the so- called “product formula”! it follows immediately that [as the notation suggests] and, in particular, is “μ log  (L )” depends only on the isomorphism class of L independent of the choice of ζ, ζ 0 . Finally, by applying the equivalences of categories of Definition 5.3, (ii), (iii), it follows immediately that we may extend the R  V - valued function [on isomorphism classes of -line bundles on M  ] μ log  (−) to a function that is also defined on isomorphism classes of -line bundles on M  , for arbitrary T {TF, TM, TLG}. Remark 5.9.1. Just as in Remark 5.3.1, one may define in the style of Corollary 5.2 a category An  [Th T , μ], where { , }, whose objects are data of the form M •μ T (Π) = (M T (Π), def log log {(R v , μ log v v )(−))} v∈V non , {(R v , μ v (X v )(−), μ̆ v (X v )(−))} v∈V arc ) where the “(Π v )’s”, “(X v )’s” preceding the “(−)’s” are to be understood as denoting the log-volumes associated, as in Definition 5.9, (ii), to the various con- stituent data of M T (Π) for Π Ob(EA  ), and whose morphisms are the mor- phisms induced by morphisms of EA  . In a similar vein, by combining the data that constitutes an object of An  [Th  T ] with the data log log {(R v , μ log v v )(−))} v∈V , {(R v , μ v (X v )(−), μ̆ v (X v )(−))} v∈V arc TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 145 where the “(Π v )’s”, “(X v )’s” preceding the “(−)’s” are to be understood as denoting the log-volumes associated, as in Definition 5.9, (ii), to the various con- stituent data of the original object of An  [Th  T ] and considering the morphisms induced by morphisms of Th  , we obtain a category An  [Th  T , μ]. Finally, by com- bining the constructions of Definitions 5.3, 5.9; Remark 5.3.1, we obtain a category An  [Th  T , ||, μ] whose objects are data of the form | M T | | def (Π) = (M T (Π), R  V , log  log  log  log v (Π)(−) } v∈V non , v (Π)(−) , μ̆ v (Π)(−) } v∈V arc , μ  (Π)(−)) where the “(Π)’s” preceding the “(−)’s” are to be understood as denoting the log-volumes associated, as in Definition 5.9, (ii), (iii), to the various constituent | | data of M T (Π) for Π Ob(EA  ), and whose morphisms are the morphisms induced by morphisms of EA  . Then, just as in Corollary 5.2, Remark 5.3.1, one obtains sequences of natural functors EA  An  [Th T , μ] An  [Th T ] Th T Th     Th  An  [Th  T , μ] An [Th T ] Th T Th     EA  An  [Th  T , ||, μ] An [Th T , ||] Th T EA where the first arrows are the functors arising from the definitions of the cat- egories “An  [−, μ]”, “An  [−, μ]”; with the exception of the second to last arrow of the first line of the above display in the case where = , every arrow of the above display is an equivalences of categories [cf. Corollary 5.2, (i), (iv), (v), (vii); Remark 5.3.1]. Remark 5.9.2. The significance of measuring [log-]volumes in units that belong to the copies of R determined by “R non (−)”, “R arc (−)” lies in the fact that such measurements may compared on both sides of the “log-wall”, as well as in a fashion compatible with the operation of mono-analyticization [cf. the discussion of Remark 3.7.7; Corollary 5.10, (ii), (iv), below]. We are now ready to state the main result of the present §5 [and, indeed, of the present paper!]. Corollary 5.10. (Fundamental Properties of Log-shells) In the notation of Corollary 5.5; Proposition 5.8, (vii), write E def = Th [Z]; An [N def ] =  An [N v ] v∈V(F mod ) where the product is a fibered product of categories over E = Th [Z]. Consider 146 SHINICHI MOCHIZUKI the diagram of categories D N v  N v   ... ... N v   ... N v  N v  ... ...  N v   ... ... ... E  An [N  ] E where we regard the rows of D as being indexed by the integers 3, 4, 5, 6, 7 [relative to which we shall use the notation “D ≤n cf. Corollary 5.5]; the ar- rows of D ≤5 are those discussed in Definition 5.6, (iii), (iv); the arrows of the rows numbered 5, 6, 7 of D are the arrows deterimined by the equivalence of cat- egories of Proposition 5.8, (vii). Note, moreover, that we have a natural mono- analyticization morphism [consisting of arrows between corresponding vertices belonging to rows indexed by the same integer!] of diagrams of categories →D D ≥3 [cf. Definition 5.4, (iv), (vi), as well as the discussion, involving panalocalization and mono-analyticization functors, of Corollary 5.5, (vi); Definition 5.6, (ii), (iii), (iv)] where the subscript “≥ 3” refers to the portion involving the rows numbered 3, 4, 5, 6, 7, and we take the arrow An [X ] An [N ] to be the arrow induced, via the equivalence of categories κ An of Corollary 5.5 and the equivalence of categories of Proposition 5.8, (vii), by the mono-analyticization functor E E ; write D for the diagram of categories obtained by gluing D , D via this mono-analyticization morphism. We shall refer to the various isomorphisms between composites of func- D tors inherent in the definition of the mono-analyticization morphism D ≥3 [e.g., the natural isomorphisms between the functors associated to the two length 2 paths N v N v N v , N v N v N v , where v V(F mod ), in the third and fourth rows of D ] as mono-analyticization homotopies. We shall refer to the natural transformation “ι v, of Corollary 5.5, (iii), as a shell-homotopy [at v] if  is a shell-arrow [cf. Definition 5.4, (iii), (v)]; we shall refer to “ι v, as a log- homotopy [at v] if the initial vertex of  is a post-log vertex. If v V(F mod ) non (respectively, v V(F mod ) arc ), then we shall refer to as a •-shell-container struc- ture on an object S Ob(N v ) the datum of an object S  Ob(X ) together with an isomorphism S λ v,ν (S  ), where ν is the terminal (respectively, initial) vertex of TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 147 the shell-arrow of  Γ × v ; an object of N v equipped with a •-shell-container structure will be referred to as a •-shell-container. Note that the shell-homotopies deter- mine •-log-shells “I” [cf. Definition 5.4, (iii), (v)] inside the underlying object of TS (respectively, TH) determined by each •-shell-container. If v V(F mod ) non (respectively, v V(F mod ) arc ), and S is an object of N v or N v , then we shall write S Gal for the topological submodule of Galois-invariants of (respectively, the topological submodule constituted by) the underlying object of TS (respectively, TH or TB) determined by S. (i) (Finite Log-volume) Let v V(F mod ) non (respectively, v V(F mod ) arc ). For each •-shell-container S Ob(N v ), S Gal is equipped with a well-defined log- volume (respectively, well-defined radial and angular log-volumes) [cf. Defini- tion 5.9, (ii)] that depend(s) only on the •-shell-container structure of S. Moreover, the •-log-shell is contained in S Gal and [relative to these log-volumes] is of finite log-volume (respectively, finite radial log-volume). (ii) (Log-Frobenius Compatibility of Log-volumes) For v, S as in (i), the log-volume (respectively, radial log-volume), computed “at  L”, is com- patible [cf. part (c) of Proposition 5.7, (i), (ii)], relative to the relevant log- homotopy, with the log-volume (respectively, angular log-volume), computed “at  + 1 L”. (iii) (Panalocalization) The log-volumes of (i), as well as the construction of the •-log-shells from the various shell-homotopies, are compatible with the panalocalization morphism D  D  of Corollary 5.5, (vi). (iv) (Mono-analyticization) If v V(F mod ) non (respectively, v V(F mod ) arc ), then we shall refer to as a •-shell-container structure on an object S Ob(N v ) the datum of an object S  Ob(X ) together with an isomorphism from S to the image in N v of λ v,ν (S  ), where ν is the terminal (respectively, initial) vertex of equipped with a •-shell-container struc- the shell-arrow of  Γ × v ; an object of N v ture will be referred to as a •-shell-container. Note that the shell-homotopies determine •-log-shells “I” [cf. Definition 5.4, (iii), (v)] inside each •-shell- container, as well as a well-defined log-volume (respectively, well-defined radial and angular log-volumes) on the Gal-superscripted module associated to a •- shell-container [cf. (i)]. These •-log-shells and log-volumes depend only on the mono-analyticized data [i.e., roughly speaking, the data contained in D ], in the following sense [cf., especially, (d)]: admits a natural structure (a) (Mono-analytic Cores) For n = 5, 6, 7, D ≤n of core on the subdiagram of categories of D determined by the union D ≤n i.e., loosely speaking, E , An [N ] “form cores” of the D ≤n−1 functors in D . (b) (Mono-analytic Telecores) As v ranges over the elements of V(F mod ) and ν over the elements of  Γ × v , the restrictions  An φ v,ν : An [N ] N v  An to An [N ] of the “forgetful functors” ψ v,ν of Proposition 5.8, D ≤6 , whose (vii), give rise to a telecore structure T An  on D ≤5 148 SHINICHI MOCHIZUKI underlying diagram of categories we denote by D An  , by appending to An  telecore edges corresponding to the arrows φ v,ν from the core D ≤6 An [N ] to the vertices of the row of D indexed by the integer 3. More- over, the respective family of homotopies of T An  and the observables S log , S log of Corollary 5.5, (iii), are compatible. (c) (Mono-analytic Contact Structures) For v V(F mod ), ν  Γ × v , there is a natural isomorphism η v,ν from the composite functor deter- 1 [of length 6] mined by the path γ v,ν X λ  v,ν −→ N v −→ N v −→ E  E  −→ An [N ] An  φ v,ν −→ N v on  Γ D An  where the first three arrows lie in D ≤5 , the fourth arrow arises from the mono-analyticization morphism D ≥3 D , and the fifth arrow lies in D to the composite functor determined by the path 0 [of length 2] γ v,ν λ  v,ν X −→ N v −→ N v on  Γ D An  . Moreover, the resulting homotopies η v,ν , v,ν ) −1 , together with the mono-analyticization homotopies and the homotopies on  [cf. Proposition 5.8, (vii)], generate a D An  arising from the “ι An v, contact structure H An  on T An  that is compatible with the telecore and contact structures T An , H An of Corollary 5.5, (ii), as well as with the homotopies of the observables S log , S log of Corollary 5.5, (iii), that  log arise from the “ι v, ”, “ι v, indexed by   Γ × v [not Γ v !]. (d) (Mono-analytic Log-shells) If v V(F mod ) non (respectively, v V(F mod ) arc ), then we shall refer to as a -shell-container structure on an object S Ob(N v ) the datum of an object S  Ob(An [N ]), An  together with an isomorphism from S φ v,ν (S  ), where ν is the ter- minal (respectively, initial) vertex of the shell-arrow of  Γ × v ; an object of N v equipped with a -shell-container structure will be referred to as a -shell-container. Note that the portion of the data that constitutes an object of An [N v ] determined by the shell-arrow gives rise to a -log- shell “I” inside each -shell-container, as well as to a log-volume on the Gal-superscripted module associated to a -shell-container [cf. Proposition 5.8, (ii), (iii), (v), (vi)]. Finally, every •-shell-container structure on an object of N v determines, by applying the isomorphism η v,ν of (c), a corresponding -shell-container structure on the object; this correspon- dence between •-, -shell-container structures is compatible with the •-, -log-shells, as well as with the various log-volumes, determined, respectively, by these •-, -shell-container structures. TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 149 Proof. The various assertions of Corollary 5.10 are immediate from the definitions, together with the references quoted in the statement of Corollary 5.10. Here, we note that in the nonarchimedean portion of part (c) of assertion (iv), in order to construct the isomorphisms η v,ν , it is necessary to relate the construction of the base field as a subset of abelianizations of various open subgroups of the Galois group [cf. Proposition 5.8, (i)] to the Kummer-theoretic construction of the base field as performed in Theorem 1.9, (e). This may be achieved by applying the group-theoretic construction algorithms of Corollary 1.10 i.e., more precisely, by combining the “fundamental class” natural isomorphism of Corollary 1.10, (a) [cf. also the first isomorphism of the display of Corollary 1.10, (b)], with the cyclotomic natural isomorphism of Corollary 1.10, (c) [cf. also Remark 1.10.3, (ii)]. Put another way, this series of algorithms may be summarized as a “group-theoretic algorithm for constructing the reciprocity map of local class field theory”. Remark 5.10.1. (i) Note that, in the notation of Corollary 5.10, (iv), (c), by pre-composing η v,ν with the telecore arrow φ  : An [X ] X of Corollary 5.5, (ii), and applying the coricity of Corollary 5.5, (i), together with an appropriate mono-analyticization homotopy, we obtain that one may think of η v,ν as yielding a homotopy from the   path φ An v,ν An [X ] −→ An [N ] −→ N v which is somewhat simpler [hence perhaps easier to grasp intuitively] than the domain path of the original homotopy η v,ν to the path An [X ] φ  −→ X λ  v,ν −→ N v −→ N v 0 [i.e., obtained by simply pre-composing γ v,ν with φ  ]. (ii) Note that the isomorphism of (i) between the two composites of functors An [X ] N v depends only on “Galois-theoretic/Aut-holomorphic data”. In particular, one may construct in the style of Remarks 5.3.1, 5.9.1 a category “An [X , η ]” whose objects consist of the data of objects of An [X ], together with the algorithms used to construct the various homotopies of (i) arising from η v,ν [i.e., associated to the various v V(F mod ), ν  Γ × v ], and whose morphisms are the morphisms induced by morphisms of An [X ]. That is to say, objects of An [X , η ] consist of objects of An [X ], together with “group-theoretic algorithms encoding the reciprocity law of local class field theory at the nonarchimedean primes and the archimedean analogue of these algorithms at the archimedean primes”. Moreover, the “forgetful functor” An [X , η ] An [X ] determines a natural equivalence of categories. Finally, one verifies immediately that one may replace “An [X ]” by “An [X , η ]” in Corollaries 5.5 and 5.10 with- out affecting the validity of their content e.g., without affecting the coricity of Corollary 5.5, (i). We leave the routine details to the interested reader. Remark 5.10.2. The significance of the theory of log-shells as summarized in Corollary 5.10 and, more generally, of the entire theory of the present paper may be understood in more intuitive terms as follows. 150 SHINICHI MOCHIZUKI (i) One important aspect of the classical theory of line bundles on a proper curve [over a field] is that although such line bundles exhibit a certain rigidity arising from the properness of the curve, this rigidity is obliterated by Zariski localization on the curve. Put another way, to work with line bundles up to isomorphism amounts to allowing oneself to “multiply the line bundle by a rational function”, i.e., to work up to rational equivalence. Although rational equivalence does not obliterate the global degree of a line bundle over the entire proper curve, if one thinks of a line bundle as a collection of integral structures at the various primes of the curve, then rational equivalence has the effect of “rearranging these integral structures” at the various primes.     v 1               v 2 v 3 v 4 v 5 ... rational equivalence       v 1 ...              v 2 v 3 v 4 v 5    ...    ... v 1               v 2 v 3 v 4 v 5 panalocalized log-shells  v 1               v 2 v 3 v 4 v 5 If one restricts oneself to working globally on the proper curve, then such “rear- rangements” are coordinated with one another in such a way as to preserve, for instance, the global degree; on the other hand, if one further imposes the condition of compatibility with Zariski localization, then such “coordination of integral struc- ture” mechanisms are obliterated. By contrast, the “MF -objects” of [Falt] satisfy a certain “extraordinary rigidity” with respect to Zariski localization that reflects the fact that they form a category that is equivalent to a certain category of Galois representations. From the point of view of thinking of line bundles as collections of integral structures at the various primes, the rigidity of MF -objects may be thought of as a sort of “freezing of the integral structures” at the various primes in a fashion that is immune to the gluing indeterminacies that occur for line bundles upon execution of Zariski localization operations. Put another way, this rigidity may be thought of as a sort of “immunity to social isolation” from other primes. In the context of Corollary 5.10, this property corresponds to the panalocalizability [i.e., Corollary 5.10, (iii)] of [the integral structures constituted by] log-shells. TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 151 (ii) At this point, it is useful to observe that, at least from an a priori point of view, there exist other ways in which one might attempt to “freeze the local integral structures”. For instance, instead of working strictly with line bundles, one could consider the ring structure of the global ring of integers of a number field [which gives rise to the trivial line bundle cf. Definition 5.3, (ii)]; that is to say, by con- sidering ring structures, one obtains a “rigid integral structure” that is compatible with Zariski localization i.e., by considering the ring structure “O” of the local rings of integers [cf. Remark 5.4.3]. Indeed, log-shells may be thought of and, moreover, were originally intended by the author as a sort of approximation of these local integral structures “O” [cf. Remark 5.4.2]. On the other hand, this sort of rigidification of local integral structures that makes essential use of the ring structure is no longer compatible with the operation of mono-analyticization [cf. Remark 5.6.1], i.e., of forgetting one of the two combinatorial dimensions “”, “” that constitute the ring structure. Thus, another crucial property of log-shells is their compatibility with mono-analyticization, as documented in Corollary 5.10, (iv) [cf. also Remarks 5.8.1, 5.9.2; Definition 5.9, (iii)], i.e., their “immunity to social isolation” from the given ring structures. From the point of view of the theory of §1, §2, §3, §4, such ring structures may be thought of as “arithmetic holomorphic structures” [i.e., outer Galois actions at nonarchimedean primes and Aut-holomorphic structures at archimedean primes] cf. Remark 5.6.1. Thus, if one thinks of the result of forgetting such “arithmetic holomorphic structures” as being like a sort of “arithmetic real analytic core” on which various “arithmetic holomorphic structures” may be imposed i.e., a sort of arithmetic analogue of the underlying real analytic surface of a Riemann surface, on which various holo- morphic structures may be imposed [cf. Remark 5.10.3 below] then the theory of mono-analyticization of log-shells guarantees that log-shells remain meaningful even as one travels back and forth between various “zones of arithmetic holo- morphy” joined in a fashion reminiscent of spokes emanating from a core by a single “mono-analytic core”. ... arith. hol. str. A ... arith. hol. str. B mono-analytic core arith. hol. str. C ... arith. hol. str. D ... 152 SHINICHI MOCHIZUKI (iii) Another approach to constructing “mono-analytic rigid local integral struc- tures” is to work with the local monoids “O  [i.e., as opposed to “log(O × )”, as was done in the case of log-shells]. Here, “O  may be thought of as a [possibly twisted] product of “O × with some “valuation monoid” that consists of a sub- monoid of R ≥0 . For instance, in the [complex] archimedean case, O C  = O C × × R >0 . On the other hand [cf. Remark 5.6.1], the dimension constituted by the “valua- tion monoid” R >0 fails to retain its rigidity when subjected to the operation of mono-analyticization. The resulting “dilations” of R >0 [i.e., by raising to the λ- th power, for λ R >0 ] may be thought of as being like Teichmüller dilations of the mono-analytic core discussed in (ii) above [cf. also the discussion of Remark 5.10.3 below]. If, moreover, one is to retain a coherent theory of global degrees of arithmetic line bundles in the presence of such “arithmetic Teichmüller dila- tions”, then [in order to preserve the “product formula” of elementary number theory] it is necessary to subject the valuation monoids at nonarchimedean primes to “arithmetic Teichmüller dilations” which are “synchronized” with the dilations that occur at the archimedean primes. From the point of the theory of Frobenioids of [Mzk16], [Mzk17], such “arithmetic Teichmüller dilations” at nonarchimedean primes are given by the unit-linear Frobenius functor studied in [Mzk16], Proposi- tion 2.5. Thus, in summary: In order to guarantee the rigidity of the local integral structures under consideration when subject to mono-analyticization, one must abandon the “valuation monoid” portion of “O  ”, i.e., one is obliged to restrict one’s attention to the “O × portion of “O  ”. On the other hand, within each zone of arithmetic holomorphy [cf. (ii)], one wishes to consider various diverse modifications of integral structure on the “rigid standard integral structures” that one constructs. Since this is not possible if one restricts oneself to “O × regarded multiplicatively, one is thus led to working with “log(O × )” i.e., in effect with the log-shells discussed in Corollary 5.10. Thus, within each zone of arithmetic holomorphy, one wishes to convert the “” operation of “O × into a “” operation, i.e., by applying the logarithm. On the other hand, when one leaves that zone of arithmetic holomorphy, one wishes to return again to work- ing with “O × multiplicatively, so as to achieve compatibility with the operation of mono-analyticization. Here, we note that -line bundles i.e., in other words, line bundles regarded from an idèlic point of view have the virtue of being defined using only the multiplicative structure of the rings involved [cf. the theory of Frobe- nioids of [Mzk16], [Mzk17]], hence of being compatible with mono-analyticization. [We remark here that the detailed specification of precisely which monoids we wish to use when we apply the theory of Frobenioids is beyond the scope of the present paper.] By contrast, although -line bundles i.e., line bundes regarded as mod- ules of a certain type are not compatible with mono-analyticization, they have the virtue of allowing us to relate, within each zone of arithmetic holomorphy, the additive module “log(O × )” to the theory of -line bundles [which is compatible with mono-analyticization]. Thus, in summary: This state of affairs obliges one to work in a “framework” in which one may pass freely, within each zone of arithmetic holomorphy, back and forth between “” and “” via application of the logarithm at the various nonarchimedean and archimedean primes. TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 153 On the other hand, since the logarithm is not a ring homomorphism, it is not at all clear, a priori, how to establish a framework in which one may apply the logarithm at will [within each zone of arithmetic holomorphy], without obliterating the foundations [e.g., scheme-theoretic!] underlying the mathematical objects that one works with, and, moreover, [a related issue cf. Remark 5.4.1] without obliterating the crucial global structure of the number fields involved [which is necessary to make sense of global arithmetic line bundles!]. A solution to this problem of finding an appropriate “framework” as dis- cussed above is precisely what is provided by “Galois theory” [cf. also the “log-invariant log-volumes” of Corollary 5.10, (i), (ii)] which is both global and “log-invariant”; the sufficiency of this “framework” [from the point of view of carrying out various arithmetic operations involving line bundles, as discussed above] is precisely what is guaranteed by the mono- anabelian theory of Corollaries 3.6, 4.5, 5.5.    log   log  core: some fixed arithmetic holomorphic structure   log !    log ... At a more philosophical level, the “log-invariant core” furnished by “Galois theory” [cf. the remarks concerning telecores following Corollaries 3.6, 3.7] and supported, in content, by “mono-anabelian geometry” may be thought of as a “geometry over F 1 [i.e., over the fictitious field of absolute constants in Z] with respect to which the logarithm is “F 1 -linear”. (iv) Note that in order to work with -line bundles [cf. the discussion of (iii); Definition 5.3, (ii)], it is necessary [unlike the case with -line bundles] to work with all the primes of a number field. Indeed, to work with “line bundles” in a fashion that allows one to ignore some nonempty set of primes of the number field amounts to working with a notion of rational equivalence that involves some proper subgroup of the multiplicative group F × associated to the number field F . On the other, the only subgroups of F × that [if one considers the union of F × with {0}] are closed under addition are the subgroups of F × that arise from subfields of F , i.e., which correspond, in effect, to -line bundles as in Definition 5.3, (ii). (v) The importance of the process of mono-analyticization in the discussion of (ii), (iii) is reminiscent of the discussion in [Mzk18], Remark 1.10.4, concerning the topic of “restricting oneself to working only with multiplicative structures” in the context of the theory of the étale theta function. 154 SHINICHI MOCHIZUKI (vi) Finally, we recall that from the point of view of the discussion of telecores in the remarks following Corollaries 3.6, 3.7, the various “forgetful functors” of assertion (ii) of Corollaries 3.6, 4.5, 5.5 may be thought of as being analogous to passing to the “underlying vector bundle plus Hodge filtration” of an MF -object [cf. Remark 3.7.2]. From this point of view: Log-shells may be thought of, in the context of this analogy with MF - objects, as corresponding to the section of a [projective] nilpotent ad- missible indigenous bundle in positive characteristic determined by the p-curvature [i.e., in other words, the Frobenius conjugate of the Hodge filtration]. Remark 5.10.3. From the point of the view of the analogy of the theory of mono- anabelian log-compatibility [cf. §3, §4] with the theory of uniformizing MF -objects [cf. Remark 3.7.2], the global/panalocal/mono-analytic theory of log-shells presented in the present §5 may be understood as follows. (i) The mathematical apparatus on a number field arising from the global/pana- local mono-anabelian log-compatibility of Corollary 5.5 may be thought of as being analogous to the [mod p] MF -object constituted by a nilpotent indigenous bundle on a hyperbolic curve in positive characteristic [cf. the theory of [Mzk1], [Mzk4]]. Note that this mathematical apparatus on a number field arises, essentially, from the outer Galois representation determined by a once-punctured elliptic curve over the number field. That is to say, roughly speaking, we have correspondences as follows: number field F once-punctured ell. curve X over F ←→ hyperbolic curve C in pos. char. ←→ nilp. indig. bundle P over C. Here, we note that the correspondence between number fields and curves over finite fields is quite classical; the correspondence between families of elliptic curves and indigenous bundles is natural in the sense that the most fundamental example of an indigenous bundle is given by the projectivization of the first de Rham cohomology module of the tautological family of elliptic curves over the moduli stack of elliptic curves. Note, moreover, that: Just as in the case of indigenous bundles, the fact that the Kodaira-Spencer morphism is an isomorphism may be interpreted as asserting that the base curve “entrusts its local moduli to the indigenous bundle”, in the mono-anabelian theory of the present paper, the various localizations of a number field “entrust their ring structures to the mono-anabelian data determined by the once-punctured elliptic curve” [cf. Remarks 1.9.4, 2.7.3, 5.6.1; Remark 5.10.2, (iii)]. Relative to this analogy, we observe that panalocalizability corresponds to the local rigidity of MF -objects [cf. Remark 5.10.2, (i)]. Moreover, the operation of mono- analyticization i.e., “forgetting the once-punctured elliptic curve” corresponds to forgetting the indigenous bundle, hence to relinquishing control of the local moduli of the base curve C; thus, just as this led to “Teichmüller dilations” in the discussion of Remark 5.10.2, (ii), (iii), in the theory of indigenous bundles, forgetting the TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 155 indigenous bundle means, in particular, loss of control of the deformation moduli of the base curve C. Another noteworthy aspect of this analogy may be seen in the fact that: Just as the log-Frobenius operation only exists for local fields [cf. Re- mark 5.4.1], in the theory of indigenous bundles, Frobenius liftings only exist Zariski locally on the base curve C. On the other hand, unlike the “linear algebra-theoretic” nature of the theory of indigenous bundles [which may be thought of as sl 2 -bundles], the outer Galois representations that appear in the theory of the present paper are fundamentally “anabelian” in nature i.e., their “non-abelian nature” is not limited to a rela- tively weak “linear algebra-theoretic” departure from abelianity, but rather on a par with that of [profinite] free groups. In particular, unlike the linear algebra- theoretic [i.e., “sl 2 -theoretic”] nature of the intertwining of the two dimensions of an indigenous vector bundle, the two combinatorial dimensions involved [cf. Re- mark 5.6.1] are intertwined in an essentially anabelian fashion [i.e., constitute a sort of “noncommutative plane”]. (ii) Once one has the “rigid standard integral structures” constituted by log- shells [cf. Remark 5.10.2, (iii)], it is natural to consider modifying these integral structures by means of the “Gaussian zeroes” [i.e., the inverse of the “Gaussian poles”] that appear in the Hodge-Arakelov theory of elliptic curves [cf., e.g., [Mzk6], §1.1]. From the point of view of this theory, this amounts, in effect, to considering the “crystalline theta object” [cf. [Mzk7], §2]. That is to say, the mathematical apparatus developed in the present §5 may be thought of as a sort of preparatory step, relative to the goal of constructing a “global MF -object- type version of the crystalline theta object”. This point of view is in line with the point of view of the Introduction to [Mzk18] [cf. also [Mzk18], Remark 5.10.2], together with the fact that the theory of the étale theta function given in [Mzk18], §1, involves, in an essential way, the theory of elliptic cuspidalizations [cf. Remark 2.7.2]. Moreover, this point of view is reminiscent of the discussion in [Mzk7], §2, of the relation of crystalline theta objects to MF -objects that is to say, the crystalline theta object has many properties that are similar to those of an MF - object, with the notable exception constituted by the vanishing of the higher p- curvatures despite the fact that the Kodaira-Spencer morphism is an isomorphism [cf. [Mzk7], Remark 2.11]. This vanishing of higher p-curvatures, when viewed from the point of view of the theory of “VF-patterns” of indigenous bundles in [Mzk4], seems to suggest that, whereas the indigenous bundles considered in the p-adic uniformization theory of [Mzk4] are of “finite Frobenius period” [in the sense that they are fixed, up to isomorphism, by some finite number of applications of Frobenius], the crystalline theta object may only be equipped with an “MF - object structure” if one allows for infinite Frobenius periods. On the other hand, by comparison to the Frobenius morphisms that appear in the theory of [Mzk4], the log-Frobenius operation log certainly has the feel of an operation of “infinite order”. Moreover, as discussed in Remark 3.6.5, the telecoricity of the mathematical apparatus of Corollary 5.5 may be regarded as being analogous to nilpotent, but non-vanishing p-curvature. That is to say: By considering the crystalline theta object not in the scheme-theoretic framework of [scheme-theoretic!] Hodge-Arakelov theory, but rather in 156 SHINICHI MOCHIZUKI the mono-anabelian framework of the present paper, one obtains a theory in which the “contradiction” [from the point of view of the classical theory of MF -objects] of “vanishing higher p-curvatures in the presence of a Kodaira-Spencer isomorphism” is naturally resolved. The above discussion suggests that one may refine the correspondence between “once-punctured elliptic curves” and “indigenous bundles” discussed in (i) as fol- lows: crystalline theta objects in scheme-theoretic Hodge-Arakelov theory the theory of mono-anabelian log-Frobenius compatibility of the present paper i.e., in essence, Belyi cuspidalization ←→ the scheme-theoretic aspects of indigenous bundles [cf. [Mzk1], Chapter I] ←→ the positive characteristic Frobenius-theoretic aspects of indigenous bundles e.g., the Verschiebung on ind. buns. [cf. [Mzk1], Chapter II] Note that the mono-anabelian theory of the present paper depends, in an essential way, on the technique of Belyi cuspidalization [cf. §1]. Since the technique of elliptic cuspidalization [cf., e.g., the theory of [Mzk18], §1!] may be thought of as a sort of simplified, linearized [cf. (v) below] version of the technique of Belyi cuspidal- ization, and the Frobenius action on square differentials in the theory of [Mzk1], Chapter II, may be identified with the derivative [i.e., a sort of “simplified, linearized version”] of the Verschiebung on indigenous bundles, it is natural to supplement the correspondences given above with the following further correspondence: the theory of the étale theta function given in [Mzk18] i.e., in essence, elliptic cuspidalization the Frobenius action on the linear space of square differentials [cf. [Mzk1], Chapter II] ←→ These analogies with the theory of [Mzk1], Chapter II, suggest the following further possible correspondences: ? hyp. orbicurves of strictly Belyi type ←→ nilp. admissible ind. buns. elliptically admissible hyp. orbicurves ←→ ? nilp. ordinary ind. buns. [i.e., where all of the hyperbolic orbicurves involved are defined over number fields cf. Remark 2.8.3]. At any rate, the correspondence with the theory of Chapters I, II of [Mzk1] suggests strongly the existence of a theory of canonical liftings for number fields equipped with a once-punctured elliptic curve that is analogous to the theory of Chapter III of [Mzk1]. The author hopes to develop such a theory in a future paper. (iii) Relative to the discussion of “units” versus “valuation monoids” in Re- mark 5.10.2, (iii), the fact that the logarithm [i.e., log-Frobenius] has the effect of TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 157 converting [a certain portion of] the “units” into a “new log-generation valuation monoid” is very much in line with the “positive slope” i.e., “telecore-theoretic” nature of a uniformizing MF -object [cf. the discussion of (i), (ii)]. Indeed, from the point of view of uniformizations of a Tate curve [cf. the discussion of Remark 2.7.2; the discussion of the Introduction of [Mzk18]] the valuation monoid portion of an MLF corresponds precisely to “slope zero”, whereas the units of an MLF correspond to “positive slope”; a similar such correspondence also appears in classical formulations of local class field theory. (iv) One important aspect of the theory of the present paper is that it is only applicable to elliptically admissible hyperbolic orbicurves, i.e., hyperbolic or- bicurves that are closely related to a once-punctured elliptic curve. In light of the “entrusting of local moduli/ring structure” aspect of the theory of the present paper discussed in (i) above, it seems reasonable to suspect that this special nature of once-punctured elliptic curves [i.e., relative to the theory of the present pa- per] may be closely related to the fact that, unlike arbitrary hyperbolic orbicurves, the moduli stack of once-punctured elliptic curves has precisely one [holomorphic] dimension [i.e., corresponding to the “one holomorphic dimension” of a number field]. This “special nature of once-punctured elliptic curves” is also reminiscent of the observation made in [Mzk6], §1.5.2, to the effect that it does not appear possible [at least in any immediate way] to generalize the scheme-theoretic Hodge- Arakelov theory of elliptic curves either to higher-dimensional abelian varieties or to higher genus curves. Moreover, it is reminiscent of the parallelogram-theoretic reconstruction algorithms of Corollary 2.7, which, from the point of view of the the- ory of [Mzk14], §2, may only be performed canonically once one chooses some fixed “one-dimensional space of square differentials” a choice which is not necessary in the elliptically admissible case, precisely because of the one-dimensionality of the moduli of once-punctured elliptic curves. (v) Observe that the “arithmetic Teichmüller dilations” discussed in Remark 5.10.2, (iii) which deform the “arithmetic holomorphic structure” are linear in nature [cf., e.g., the “unit-linear Frobenius functor”]. On the other hand, the log- Frobenius operation within each “zone of arithmetic holomorphy” is “non-linear”, with respect to both the additive and multiplicative structures of the rings involved. Indeed, as discussed extensively in the remarks following Corollaries 3.6, 3.7 [cf. also the discussion in the latter half of Remark 5.10.2, (iii)], the essential reason for the introduction of mono-anabelian geometry in the present paper is precisely the need to deal with this non-linearity. In the classical theory of Teichmüller deformations of Riemann surfaces, the deformations of holomorphic structure are linear [cf. the approach to this theory given in [Mzk14], §2]. On the other hand, non-linearity may be witnessed in classical Teichmüller theory in the quadratic nature of the square differentials. Typically, non-linearity is related to some sort of “bounded domain”. In the complex theory, the bounded nature of the upper half-plane, as well as of Teichmüller space itself, constitute examples of this phenomenon cf. the discussion of “Frobenius-invariant integral structures” in [Mzk4], Introduction, §0.4. In the case of elliptic curves, the quadratic nature of the square differentials corresponds precisely to the quadratic nature of the exponent that appears in the classical series representation of the theta function; moreover, this quadratic cor- respondence “Z n → n 2 Z” is [unlike the linear correspondence n → c · n, for c Z] bounded from below. Returning to the theory of log-shells, let us recall 158 SHINICHI MOCHIZUKI that the non-linear log-Frobenius operation is used precisely to achieve the cru- cial boundedness [i.e., “compactness”] property of log-shells [cf. the discussion of Remark 5.10.2!]. Also, relative to the discussion of (ii) above, let us recall that the goal of constructing a comparison isomorphism between non-linear compact domains of function spaces formed one of the key motivations for the development of the Hodge-Arakelov theory of elliptic curves [cf. [Mzk6], §1.3.2, §1.3.3]. (vi) Relative to the analogy between “once-punctured elliptic curves over num- ber fields” and “nilpotent indigenous bundles” [cf. (i)], it is interesting to note that if one thinks of the number fields involved as “log number fields” i.e., number fields equipped with a finite set of primes at which the elliptic curve is allowed to have bad [but multiplicative!] reduction then Siegel’s classical finiteness theorem [which implies the finiteness of the set of isomorphism classes of elliptic curves over a given “log number field”] may be regarded as the analogue of the finiteness of the Verschiebung on indigenous bundles given in [Mzk1], Chapter II, Theorem 2.3 [which implies the finiteness of the set of isomorphism classes of nilpotent indigenous bundles over a given hyperbolic curve in positive characteristic]. Remark 5.10.4. The analogy with Frobenius liftings that appears in the discus- sion of Remark 5.10.3 is interesting from the point of view of the theory of [Mzk21], §2 [cf., especially [Mzk21], Remark 2.9.1]. Indeed, [Mzk21], §2, may be thought of as a theory concerning the issue of passing from decomposition groups to ring [i.e., additive!] structures in a p-adic setting [cf. [Mzk21], Corollary 2.9], hence may be thought of as a sort of p-adic analogue of the lemma of Uchida reviewed in Proposition 1.3. TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 159 Appendix: Complements on Complex Multiplication In the present Appendix, we expose the portion of the well-known theory of abelian varieties with complex multiplication [cf., e.g., [Lang-CM], [Milne-CM], for more details] that underlies the observation “(∗ CM )” i.e., roughly speaking, to the effect that, if p is a prime number, then every Lubin-Tate character on an open subgroup of the inertia group of the absolute Galois group of a p-adic local field arises, after possible restriction to an open subgroup, from a subquotient of the p-adic Tate module associated to an abelian variety with complex multiplication related to the author by A. Tamagawa [cf. [Mzk20], Remark 3.8.1]. In particular, we verify that this observation (∗ CM ) does indeed hold. [Here, we remark in passing that the proof of (∗ CM ) given in the present Appendix is, according to Tamagawa, apparently somewhat different from the proof that he originally considered. Unfor- tunately, however, he was unable to recall the details of his original argument.] This implies that the observation “(∗ A-qLT )” discussed in [Mzk20], Remark 3.8.1, also holds, and hence, in particular, that the hypothesis of [Mzk20], Corollary 3.9, to the effect that “either (∗ A-qLT ) or (∗ CM ) holds” may be eliminated [i.e., that [Mzk20], Corollary 3.9, holds unconditionally]. On the other hand, we conclude the present Appendix by observing that, in this context, there still remains an interesting open problem that could serve to stimulate further research. In the following, we shall fix a prime number p and write Q for the field of rational numbers, Z p for the topological ring of p-adic integers, Q p for the topolog- ical field of p-adic numbers, R for the topological field of real numbers, C for the topological field of complex numbers, ι : C C for the automorphism of C given by complex conjugation, and Q alg C for the subfield of algebraic numbers. Also, we shall use the notation “O” to denote the ring of integers associated to a finite extension of Q or Q p and the notation “tr (−) to denote the trace map associated to a finite field extension “(−)”. (CM1) Fix a finite extension L of degree d 1 of Q p . Thus, L = Q p (α) for some α L. Let f (x) Q p [x] be a monic irreducible polynomial such that f (α) = 0. If def def d = 2, then set g(x) = x 2 + 1; if d  = 2, then set g(x) = (x 1)(x 2) · . . . · (x d). Thus, both f (x) Q p [x] and g(x) Q[x] are of degree d. Then by approximating the coefficients of f and g by elements of Q at the p-adic and real places of Q, we conclude that there exists a monic polynomial h(x) Q[x] of degree d such that the following conditions hold: (a) there exists an element β L such that h(β) = 0 and L = Q p (β); (b) if d = 2, then the complex roots of h(x) are non-real and distinct; (c) if d  = 2, then the complex roots of h(x) are real and distinct. Indeed, (a) follows by arguing as in [Kobl], pp. 69-70; (b) follows by considering the sign of the discriminant of h(x); (c) follows by considering the signs of values of g(x) as x varies over the real numbers in the various intervals between roots of g(x). Note that it follows from (a) that the polynomial h(x) Q[x] is irreducible. Thus, we obtain a number field 160 SHINICHI MOCHIZUKI def F = Q[x]/(h(x)) such that [F : Q] = d, and F Q Q p is isomorphic to L. If d = 2, then F is a complex quadratic extension of Q, hence admits an element γ F \ Q such that γ 2 Q [which implies that γ 2 < 0, F = Q(γ)]. Next, let us observe [cf. [Kobl], p. 81] that 1 p 3 admits a square root in Q p . Thus, if d  = 2, then the number field F is totally real and hence linearly disjoint over Q from the complex quadratic def extension K 0 = Q(λ 0 ), where λ 20 = 1 p 3 . In particular, if d  = 2, then the number field def K = F · K 0 is a CM field [cf., e.g., [Lang-CM], Chapter 1, §2] of degree 2d over Q. (CM2) Suppose that d = 2. Let ϕ 0 : F → C be an embedding such that the imaginary part of ϕ 0 (γ) is positive. Write ι F Gal(F/Q) for the unique nontrivial element of Gal(F/Q) [so ϕ 0 ι F = ι ϕ 0 ]. Then recall [cf., e.g., [Lang-CM], Chapter 1, §4] that the complex torus C/ϕ 0 (O F ), together with the Riemann form determined by the pairing (ξ, η) → tr F/Q · ι F (η) · γ) Q [where ξ, η F ], determine an elliptic curve E with complex multiplication by O F , which is defined over some finite subextension M of ϕ 0 (F ) in C. Now it is immediate from the Main Theorem of Complex Multiplication [i.e., Shimura reciprocity cf., e.g., [Lang-CM], Chapter 4, Theorem 1.1; [Milne-CM], Theorem 10.1] that there def exists an open subgroup H of the inertia group G M = Gal(Q alg /M ) associated to some prime of Q alg that divides p such that H acts on the p-adic Tate module associated to E via the Lubin-Tate character associated to L. This completes the proof of (∗ CM ) in the case d = 2. (CM3) Suppose that d  = 2. Let Φ 0 be a collection of d embeddings K → C of K into the complex numbers such that every embedding F → C is obtained as the restriction of an element of Φ 0 , and, moreover, the embeddings of Φ 0 map λ 0 to a complex number whose imaginary part is positive. [Thus, the embeddings of Φ 0 coincide on K 0 .] Fix an element ϕ 0 Φ 0 . Thus, one verifies immediately that both Φ 0 and def Φ = 0 } ϕ | ϕ 0  = ϕ Φ 0 } form CM types of K [cf., e.g., [Lang-CM], Chapter 1, §2]. Moreover, if we write def Φ ι 0 = ϕ | ϕ Φ 0 }, then one verifies immediately that the set of embeddings K → C [or, equivalently, K → Q alg ] Φ 0 Φ ι 0 Φ 0 × {id, ι} def [where id denotes the identity automorphism of C] admits a natural action by G Q = Gal(Q alg /Q) that preserves the product decomposition [induced by restricting the embeddings in question to F or K 0 ] of the above display. Then one verifies immedi- def ately that the subgroup of G Q that stabilizes Φ 0 is equal to G K 0 = Gal(Q alg /K 0 ), and hence that the reflex field [cf., e.g., [Lang-CM], Chapter 1, §5] associated to (K, Φ 0 ) is equal to ϕ 0 (K 0 ). On the other hand, observe that our assumption that d  = 2 implies that the cardinalities [namely, 1 and d 1] of the intersections Φ Φ 0 and Φ Φ ι 0 are distinct. Thus, since the action of any element of G Q on Φ 0 Φ ι 0 is TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 161 compatible with the projection to the set {id, ι}, one verifies immediately [by consid- ering the fibers, i.e., Φ 0 and Φ ι 0 , of this projection] that our assumption that d  = 2 implies that an element of G Q stabilizes Φ if and only if it fixes ϕ 0 . In particular, we conclude that the reflex field [cf., e.g., [Lang-CM], Chapter 1, §5] associated to (K, Φ) is equal to ϕ 0 (K). (CM4) We continue our analysis of the situation discussed in (CM3). Write ι K Gal(K/F ) for the unique nontrivial element of Gal(K/F ) [so ϕ ι K = ι ϕ, for all ϕ Φ]. Observe that by approximating λ 0 relative to ϕ 0 and −λ 0 relative to ϕ Φ \ 0 }, one may construct an element λ K such that the imaginary part of ϕ(λ) is positive for all ϕ Φ. Moreover, by replacing λ by λ ι K (λ), one may assume without loss of generality that ι K (λ) = −λ. Next, recall [cf., e.g., [Lang- CM], Chapter 1, §4] that the CM type (K, Φ), together with the lattice O K K and the Riemann form determined by the pairing (ξ, η) → tr K/Q · ι K (η) · λ) Q [where ξ, η K], determine a polarized abelian variety A with complex multiplication by O K , which is defined over some finite subextension M of ϕ 0 (K) def in C. Next, write G M = Gal(Q alg /M ), T p (A) for the p-adic Tate module associated to A. Thus, T p (A) admits a natural structure of rank one free O K Z p -module, as well as a natural G M -action. In particular, since O K Z p = O L O L , we thus conclude that T p (A) admits a direct sum decomposition T p (A) = T  T  as a direct sum of rank one free O L -modules T  , T  . On the other hand, let us recall that the Main Theorem of Complex Multiplication [i.e., Shimura reciprocity cf., e.g., [Lang-CM], Chapter 4, Theorem 1.1; [Milne-CM], Theorem 10.1] allows one to compute the Galois action of G M on T p (A) by means of the reflex type norm applied to an idèle of M . In particular, it follows immediately from our construction of Φ from Φ 0 in (CM3), together with the resulting computation of the associated reflex field, that, after possibly interchanging T  and T  , there exists an open subgroup H of the inertia group G M associated to some prime of Q alg that divides p such that H acts on T 1 via the Lubin-Tate character × χ LT : H O L associated to L [i.e., in essence, via the embedding ϕ 0 ] and on T 2 via the dual character × χ LT : H O L [that is to say, the character determined by the relation χ LT · χ LT = χ cycl , where χ cycl : H Z × p is the cyclotomic character, i.e., in essence, via the product of the embeddings Φ \ 0 }]. This completes the proof of the observation (∗ CM ) [for arbitrary d]. (CM5) The above argument completes the proof of the observation (∗ CM ) and hence also of the observation (∗ A-qLT ), of [Mzk20], Remark 3.8.1. On the other hand, we conclude by observing that, in this context, the following problem remains unresolved: Let X be a hyperbolic curve over a finite extension k of Q p . Then is it always the case that the étale fundamental group of X is of A-qLT-type [cf. [Mzk20], Definition 3.1, (v)]? Here, we recall that, roughly speaking, this condition of being “of A-qLT-type” may be described as the condition that every Lubin-Tate character on the inertia 162 SHINICHI MOCHIZUKI subgroup of an open subgroup of the absolute Galois group of k arises, after possibly restricting to an open subgroup, from some subquotient of the p-adic Tate module of the Jacobian of a finite étale covering of X [cf. [Mzk20], Definition 3.1, (v), for more details]. Thus, (∗ A-qLT ) consists of the assertion that this problem admits an affirmative answer whenever X admits a finite étale covering that, in turn, admits a dominant map to a copy of the projective line minus three points over k. We recall from [Mzk20], Remark 3.8.1, that (∗ A-qLT ) is derived from (∗ CM ) by using Belyi maps. Thus, the above unresolved problem is particularly of interest in the case of various “classes” of X for which techniques involving Belyi maps cannot be applied, e.g., the case of proper X. Finally, we observe that this problem may also be understood in the context of the general theme of applications of Belyi maps, i.e., in the style of Belyi injectivity or [André], Theorems 7.2.1, 7.2.3 [which may be thought of as a sort of p- adic version of Belyi injectivity]. In the case of Belyi injectivity or André’s results, a version for arbitrary hyperbolic curves was obtained, by applying techniques from combinatorial anabelian ge- ometry, in [HM1], Theorem C [in the case of Belyi injectivity] and [HM2], Theorem B [in the case of André’s results]. On the other hand, in the case of the unresolved problem discussed above, it is not clear to the author at the present time how to apply techniques from combinatorial anabelian geometry to resolve this problem. TOPICS IN ABSOLUTE ANABELIAN GEOMETRY III 163 Bibliography [André] Y. André, On a Geometric Description of Gal(Q p /Q p ) and a p-adic Avatar of  , Duke Math. J. 119 (2003), pp. 1-39. GT [ArTt] E. Artin, J. Tate, Class Field Theory, Addison-Wesley (1990). [Falt] G. Faltings, Crystalline Cohomology and p-adic Galois Representations, JAMI Conference, Johns Hopkins Univ. Press (1990), 25-79. [HM1] Y. Hoshi, S. Mochizuki, On the Combinatorial Anabelian Geometry of Nodally Nondegenerate Outer Representations, Hiroshima Math. J. 41 (2011), pp. 275- 342. [HM2] Y. Hoshi, S. Mochizuki, Topics Surrounding the Combinatorial Anabelian Ge- ometry of Hyperbolic Curves III: Tripods and Tempered Fundamental Groups, RIMS Preprint 1763 (November 2012). [Ih] Y. Ihara, Lifting Curves over Finite Fields Together with the Characteristic Correspondence Π + Π  , Jour. of Algebra 75 (1982), pp. 452-483. [Kobl] N. Koblitz, p-adic Numbers, p-adic Analysis, and Zeta Functions, Graduate Texts in Mathematics 58, Springer-Verlag (1977). [Lang-CM] S. Lang, Complex Multiplication, Grundlehren der Mathematischen Wissenschaften 255, Springer-Verlag (1983). [Milne-CM] J. S. Milne, Complex Multiplication, manuscript available at the following ad- dress: http://www.jmilne.org/math/CourseNotes/cm.html [Mzk1] S. Mochizuki, A Theory of Ordinary p-adic Curves, Publ. of RIMS 32 (1996), pp. 957-1151. [Mzk2] S. Mochizuki, A Version of the Grothendieck Conjecture for p-adic Local Fields, The International Journal of Math. 8 (1997), pp. 499-506. [Mzk3] S. Mochizuki, Correspondences on Hyperbolic Curves, Journ. Pure Appl. Al- gebra 131 (1998), pp. 227-244. [Mzk4] S. Mochizuki, Foundations of p-adic Teichmüller Theory, AMS/IP Studies in Advanced Mathematics 11, American Mathematical Society/International Press (1999). [Mzk5] S. Mochizuki, The Local Pro-p Anabelian Geometry of Curves, Invent. Math. 138 (1999), pp. 319-423. [Mzk6] S. Mochizuki, A Survey of the Hodge-Arakelov Theory of Elliptic Curves I, Arithmetic Fundamental Groups and Noncommutative Algebra, Proceedings of Symposia in Pure Mathematics 70, American Mathematical Society (2002), pp. 533-569. [Mzk7] S. Mochizuki, A Survey of the Hodge-Arakelov Theory of Elliptic Curves II, Algebraic Geometry 2000, Azumino, Adv. Stud. Pure Math. 36, Math. Soc. Japan (2002), pp. 81-114. [Mzk8] S. Mochizuki, Topics Surrounding the Anabelian Geometry of Hyperbolic Curves, Galois Groups and Fundamental Groups, Mathematical Sciences Research In- stitute Publications 41, Cambridge University Press (2003), pp. 119-165. 164 SHINICHI MOCHIZUKI [Mzk9] S. Mochizuki, The Absolute Anabelian Geometry of Hyperbolic Curves, Galois Theory and Modular Forms, Kluwer Academic Publishers (2004), pp. 77-122. [Mzk10] S. Mochizuki, The Absolute Anabelian Geometry of Canonical Curves, Kazuya Kato’s fiftieth birthday, Doc. Math. 2003, Extra Vol., pp. 609-640. [Mzk11] S. Mochizuki, The Geometry of Anabelioids, Publ. Res. Inst. Math. Sci. 40 (2004), pp. 819-881. [Mzk12] S. Mochizuki, Galois Sections in Absolute Anabelian Geometry, Nagoya Math. J. 179 (2005), pp. 17-45. [Mzk13] S. Mochizuki, Semi-graphs of Anabelioids, Publ. Res. Inst. Math. Sci. 42 (2006), pp. 221-322. [Mzk14] S. Mochizuki, Conformal and Quasiconformal Categorical Representation of Hyperbolic Riemann Surfaces, Hiroshima Math. J. 36 (2006), pp. 405-441. [Mzk15] S. Mochizuki, Global Solvably Closed Anabelian Geometry, Math. J. Okayama Univ. 48 (2006), pp. 57-71. [Mzk16] S. Mochizuki, The Geometry of Frobenioids I: The General Theory, Kyushu J. Math. 62 (2008), pp. 293-400. [Mzk17] S. Mochizuki, The Geometry of Frobenioids II: Poly-Frobenioids, Kyushu J. Math. 62 (2008), pp. 401-460. [Mzk18] S. Mochizuki, The Étale Theta Function and its Frobenioid-theoretic Manifes- tations, Publ. Res. Inst. Math. Sci. 45 (2009), pp. 227-349. [Mzk19] S. Mochizuki, Absolute anabelian cuspidalizations of proper hyperbolic curves, J. Math. Kyoto Univ. 47 (2007), pp. 451-539. [Mzk20] S. Mochizuki, Topics in Absolute Anabelian Geometry I: Generalities, J. Math. Sci. Univ. Tokyo 19 (2012), pp. 139-242. [Mzk21] S. Mochizuki, Topics in Absolute Anabelian Geometry II: Decomposition Groups and Endomorphisms, J. Math. Sci. Univ. Tokyo 20 (2013), pp. 171-269. [MT] S. Mochizuki, A. Tamagawa, The algebraic and anabelian geometry of config- uration spaces, Hokkaido Math. J. 37 (2008), pp. 75-131. [NSW] J. Neukirch, A. Schmidt, K. Wingberg, Cohomology of number fields, Grundlehren der Mathematischen Wissenschaften 323, Springer-Verlag (2000). [Pop] F. Pop, On Grothendieck’s conjecture of anabelian birational geometry, Ann. of Math. 138 (1994), pp. 147-185. [Serre] J.-P. Serre, Lie Algebras and Lie Groups, Lecture Notes in Mathematics 1500, Springer Verlag (1992). [Stix] J. Stix, Affine anabelian curves in positive characteristic, Compositio Math. 134 (2002), pp. 75-85. [Tama] A. Tamagawa, The Grothendieck Conjecture for Affine Curves, Compositio Math. 109 (1997), pp. 135-194. [Uchi] K. Uchida, Isomorphisms of Galois groups of algebraic function fields, Ann. of Math. 106 (1977), pp. 589-598.